├── DESCRIPTION ├── LICENSE ├── NAMESPACE ├── NEWS.md ├── R ├── bAccuracy.R ├── data.R ├── matPCs.R └── multiAdaSampling.R ├── README.md ├── data └── gse87795_subset_sce.rda ├── img ├── scReClassify.jpg └── scReClassify_sticker.png ├── inst ├── CITATION ├── scReClassify.jpg └── scReClassify_sticker.png ├── man ├── bAccuracy.Rd ├── gse87795_subset_sce.Rd ├── matPCs.Rd ├── multiAdaSampling.Rd └── scReClassify.Rd ├── tests ├── testthat.R └── testthat │ ├── test-bAccuracy.R │ ├── test-matPCs.R │ └── test-multiAdaSampling.R └── vignettes └── scReClassify.Rmd /DESCRIPTION: -------------------------------------------------------------------------------- 1 | Package: scReClassify 2 | Type: Package 3 | Title: scReClassify: post hoc cell type classification of single-cell RNA-seq data 4 | Version: 0.99.8 5 | Authors@R: c( 6 | person("Pengyi", "Yang", email = "pengyi.yang@sydney.edu.au", 7 | role = c("aut"), comment = c(ORCID = "0000-0003-1098-3138")), 8 | person("Taiyun", "Kim", email = "taiyun.kim91@gmail.com", 9 | role = c("aut", "cre"), comment = c(ORCID = "0000-0002-5028-836X"))) 10 | Description: A post hoc cell type classification tool to fine-tune cell type 11 | annotations generated by any cell type classification procedure with 12 | semi-supervised learning algorithm AdaSampling technique. 13 | The current version of scReClassify supports Support Vector Machine and 14 | Random Forest as a base classifier. 15 | License: GPL-3 + file LICENSE 16 | BugReports: https://github.com/SydneyBioX/scReClassify/issues 17 | URL: https://github.com/SydneyBioX/scReClassify, http://www.bioconductor.org/packages/release/bioc/html/scReClassify.html 18 | Depends: R (>= 4.1) 19 | Encoding: UTF-8 20 | LazyData: false 21 | RoxygenNote: 7.1.2 22 | Roxygen: list(markdown = TRUE) 23 | Imports: 24 | randomForest, 25 | e1071, 26 | stats, 27 | SummarizedExperiment, 28 | SingleCellExperiment, 29 | methods 30 | VignetteBuilder: knitr 31 | biocViews: Software, Transcriptomics, SingleCell, Classification, SupportVectorMachine 32 | Suggests: 33 | testthat, 34 | knitr, 35 | BiocStyle, 36 | rmarkdown, 37 | DT, 38 | mclust, 39 | dplyr 40 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /NAMESPACE: -------------------------------------------------------------------------------- 1 | # Generated by roxygen2: do not edit by hand 2 | 3 | export(bAccuracy) 4 | export(matPCs) 5 | export(multiAdaSampling) 6 | import(SingleCellExperiment) 7 | importFrom(SummarizedExperiment,assay) 8 | importFrom(e1071,svm) 9 | importFrom(methods,is) 10 | importFrom(randomForest,randomForest) 11 | importFrom(stats,median) 12 | importFrom(stats,prcomp) 13 | importFrom(stats,predict) 14 | importFrom(stats,sd) 15 | -------------------------------------------------------------------------------- /NEWS.md: -------------------------------------------------------------------------------- 1 | ## scReCkassify 0.99.8 2 | * Included citation for the package 3 | 4 | ## scReCkassify 0.99.7 5 | * Removed global assignment in multiAdaSampling. 6 | 7 | ## scReCkassify 0.99.6 8 | * Revised further to address comments/feedbacks 9 | 10 | ## scReCkassify 0.99.5 11 | * Fixed NEWS to match the version bump 12 | 13 | 14 | ## scReCkassify 0.99.4 15 | * Minor change to example in `matPC` function 16 | 17 | 18 | ## scReCkassify 0.99.3 19 | * Correct for any spelling errors in all documentations 20 | * Revising the package to address the comments from Bioconductor review 21 | * Cleaned the GSE87795 subset data to a SingleCellExperiment object 22 | 23 | 24 | ## scReCkassify 0.99.2 25 | * version bump with submission to Bioconductor 26 | 27 | ## scReClassify 0.99.1 28 | * Cleaning repository to pass BiocCheck 29 | 30 | ## scReClassify 0.99.0 31 | * Initial submission for Bioconductor 32 | * Code formats/documentations have been revised to meet Bioconductor 33 | requirements 34 | 35 | ## scReClassify 0.1.1 36 | 37 | * Significant updates and addition to the documentations 38 | * Major changes to code writing style to comply with BioC 39 | * Updated example data 40 | -------------------------------------------------------------------------------- /R/bAccuracy.R: -------------------------------------------------------------------------------- 1 | #' bAccuracy 2 | #' 3 | #' This function calculates the accuracy of the prediction to the true label. 4 | #' 5 | #' @param cls.truth A character vector of true class label. 6 | #' @param final A vector of final classified label prediction from 7 | #' \code{multiAdaSampling}. 8 | #' @return An accuracy value. 9 | #' @author Pengyi Yang, Taiyun Kim 10 | #' 11 | #' @examples 12 | #' data("gse87795_subset_sce") 13 | #' 14 | #' mat.expr <- gse87795_subset_sce 15 | #' cellTypes <- gse87795_subset_sce$cellTypes 16 | #' 17 | #' # Get dimension reduced matrix. We are using `logNorm` assay from `mat.expr`. 18 | #' mat.pc <- matPCs(mat.expr, assay = "logNorm") 19 | #' 20 | #' # Here we are using Support Vector Machine as a base classifier. 21 | #' result <- multiAdaSampling(mat.pc, cellTypes, classifier = "svm", 22 | #' percent = 1, L = 10) 23 | #' 24 | #' final <- result$final 25 | #' 26 | #' # Balanced accuracy 27 | #' bacc <- bAccuracy(cellTypes, final) 28 | #' 29 | #' @export bAccuracy 30 | bAccuracy <- function(cls.truth, final) { 31 | 32 | if (length(cls.truth) != length(final)) { 33 | stop("`cls.truth` and `final` objects have different length.") 34 | } 35 | 36 | gs <- names(table(cls.truth)) 37 | acc <- c() 38 | acc <- unlist(lapply(seq_len(length(gs)), function(i) { 39 | sum((cls.truth == final)[cls.truth==gs[i]]) / sum(cls.truth==gs[i]) 40 | })) 41 | mean(acc) 42 | } 43 | -------------------------------------------------------------------------------- /R/data.R: -------------------------------------------------------------------------------- 1 | #' @title GSE827795 subset data 2 | #' 3 | #' @description A SingleCellExperiment object containing a subset expression 4 | #' matrix of GSE827795. The data contains log2 transformed FPKM expression. 5 | #' 6 | #' GSE87795 is a mouse fetal liver development data containing 1000 genes, 7 | #' 367 cells and 6 cell types. 8 | #' 9 | #' The original GSE87795 data and the study details can be found at this 10 | #' [link](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE87795) 11 | #' 12 | #' 13 | #' @name gse87795_subset_sce 14 | #' @docType data 15 | #' 16 | "gse87795_subset_sce" 17 | 18 | 19 | 20 | #' scReClassify: a package for post hoc cell type classification of single-cell 21 | #' RNA-sequencing data. 22 | #' 23 | #' @description A post hoc cell type classification tool to fine-tune cell type 24 | #' annotations generated by any cell type classification procedure with 25 | #' semi-supervised learning algorithm AdaSampling technique. 26 | #' 27 | #' The current version of scReClassify supports Support Vector Machine and 28 | #' Random Forest as a base classifier. 29 | #' 30 | #' @author 31 | #' **Maintainer**: 32 | #' 33 | #' * Taiyun Kim (ORCID:0000-0002-5028-836X) 34 | #' - Email: taiyun.kim91@gmail.com 35 | #' 36 | #' Authors: 37 | #' 38 | #' * Pengyi Yang (ORCID: 0000-0003-1098-3138) 39 | #' 40 | #' 41 | #' @seealso 42 | #' 43 | #' Useful links: 44 | #' * Vignette available at: https://sydneybiox.github.io/scdney/ 45 | #' 46 | #' @docType package 47 | #' @name scReClassify 48 | NULL 49 | 50 | -------------------------------------------------------------------------------- /R/matPCs.R: -------------------------------------------------------------------------------- 1 | #' matPCs function 2 | #' 3 | #' @description Performs PCA on a given matrix and returns a dimension reduced 4 | #' matrix which captures at least 80% (default) of overall variability. 5 | #' 6 | #' @param data An expression matrix or a SingleCellExperiment object. 7 | #' @param assay An assay to select if \code{data} is a SingleCellExperiment 8 | #' object 9 | #' @param percentVar The percentage of variance threshold. This is used to 10 | #' select number of Principal Components. 11 | #' @details This function performs PCA to reduce the dimension of the gene 12 | #' expression matrix limited from 10 to 20 PCs. 13 | #' @return Dimensionally reduced matrix. 14 | #' @author Pengyi Yang, Taiyun Kim 15 | #' @export matPCs 16 | #' @examples 17 | #' data("gse87795_subset_sce") 18 | #' 19 | #' mat.expr <- gse87795_subset_sce 20 | #' 21 | #' mat.pc <- matPCs(mat.expr, assay = "logNorm") 22 | #' 23 | #' # to capture at least 70% of overall variability in the dataset, 24 | #' mat.dim.reduct.70 <- matPCs(mat.expr, assay = "logNorm", 0.7) 25 | #' 26 | #' @importFrom stats prcomp sd 27 | #' @importFrom SummarizedExperiment assay 28 | #' @importFrom methods is 29 | #' @import SingleCellExperiment 30 | matPCs <- function(data, assay = NULL, percentVar=0.8) { 31 | 32 | if (percentVar < 0 | percentVar > 1) { 33 | stop("percentVar must be a number between 0 and 1") 34 | } 35 | # If SCE 36 | if (is(data, "SingleCellExperiment")) { 37 | if (is.null(assay)) { 38 | stop("assay parameter cannot be NULL") 39 | } 40 | mat <- SummarizedExperiment::assay(data, assay) 41 | } else { 42 | mat <- data 43 | } 44 | 45 | if (is(data, "matrix") | is(data, "data.frame")) { 46 | pca <- stats::prcomp(t(mat), center = TRUE, scale. = TRUE) 47 | } else if ((is(data, "SingleCellExperiment")) & 48 | ("PCA" %in% reducedDimNames(data))) { 49 | pca <- SingleCellExperiment::reducedDim(data, "PCA") 50 | } else if (is(data, "SingleCellExperiment")) { 51 | pca <- stats::prcomp(t(mat), center = TRUE, scale. = TRUE) 52 | if (length(SingleCellExperiment::reducedDimNames(data))) { 53 | rd <- SingleCellExperiment::reducedDims(data) 54 | rd$PCA = pca$x 55 | } else { 56 | SingleCellExperiment::reducedDims(data) <- list(PCA = pca$x) 57 | } 58 | } 59 | 60 | # genes are rows, cells are cols 61 | pcs <- c() 62 | sdev <- apply(pca$x, 2, sd) 63 | eigs <- sdev^2 64 | top <- which(cumsum(eigs)/sum(eigs) > percentVar)[1] 65 | m <- ifelse(top > 20, 20, top) 66 | m <- ifelse(m < 10, 10, m) 67 | pcs <- pca$x[,seq_len(m)] 68 | 69 | return(pcs) 70 | } 71 | -------------------------------------------------------------------------------- /R/multiAdaSampling.R: -------------------------------------------------------------------------------- 1 | #' multi Adaptive Sampling function 2 | #' 3 | #' Performs multiple adaptive sampling to train a classifier model. 4 | #' 5 | #' @param data A dimension reduced matrix from \code{matPCs}. 6 | #' @param label A vector of label information for each sample. 7 | #' @param reducedDimName A name of the \code{reducedDim} to use. This must be 8 | #' specified if \code{data} is a SingleCellExperiment object. 9 | #' @param classifier Base classifier model, either "SVM" (\code{svm}) or "RF" 10 | #' \code{'rf'} is supported. 11 | #' @param percent Percentage of samples to select at each iteration. 12 | #' @param L Number of ensembles. Default to 10. 13 | #' @param prob logical flag to return sample's probabilities to each class. 14 | #' @param balance logical flag to if the cell types are balanced. 15 | #' If `FALSE`, down sample large cell types classes to the median of all class 16 | #' sizes. 17 | #' @param iter A number of iterations to perform adaSampling. 18 | #' @return A final prediction, probabilities for each cell type and the model 19 | #' are returned as a list. 20 | #' 21 | #' @author Pengyi Yang, Taiyun Kim 22 | #' @importFrom randomForest randomForest 23 | #' @importFrom e1071 svm 24 | #' @importFrom stats median predict 25 | #' @import SingleCellExperiment 26 | #' @examples 27 | #' 28 | #' library(SingleCellExperiment) 29 | #' 30 | #' # Loading the data 31 | #' data("gse87795_subset_sce") 32 | #' 33 | #' mat.expr <- gse87795_subset_sce 34 | #' cellTypes <- gse87795_subset_sce$cellTypes 35 | #' 36 | #' # Get dimension reduced matrix. We are using `logNorm` assay from `mat.expr`. 37 | #' reducedDim(mat.expr, "matPCs") <- matPCs(mat.expr, assay = "logNorm") 38 | #' 39 | #' # Here we are using Support Vector Machine as a base classifier. 40 | #' result <- multiAdaSampling(mat.expr, cellTypes, reducedDimName = "matPCs", 41 | #' classifier = "svm", percent = 1, L = 10) 42 | #' @export multiAdaSampling 43 | multiAdaSampling <- function(data, label, reducedDimName = NULL, 44 | classifier="svm", percent=1, L=10, prob=FALSE, 45 | balance=TRUE, iter=3) { 46 | # If SCE 47 | if (is(data, "SingleCellExperiment")) { 48 | if (is.null(reducedDimName)) { 49 | stop("reducedDimName parameter cannot be NULL if data is a", 50 | " SingleCellExperiment object") 51 | } 52 | mat <- SingleCellExperiment::reducedDim(data, reducedDimName) 53 | } else { 54 | mat <- data 55 | } 56 | mat <- t(mat) 57 | 58 | if (ncol(mat) != length(label)) { 59 | stop("Parameter `label` and number of columns of data must be equal") 60 | } 61 | if (percent < 0 | percent > 1) { 62 | stop("Parameter percent cannot must be a value between 0 and 1") 63 | } 64 | 65 | models <- lapply(seq_len(L), function(l) { 66 | X <- mat 67 | Y <- label 68 | 69 | model <- c() 70 | prob.mat <- c() 71 | 72 | for (i in seq_len(iter)) { 73 | if (classifier == "rf") { 74 | model <- randomForest::randomForest(t(X), factor(Y), 75 | ntree = 100) 76 | prob.mat <- stats::predict(model, newdata=t(mat), type="prob") 77 | } 78 | if (classifier == "svm") { 79 | tmp <- t(X) 80 | rownames(tmp) <- NULL 81 | model <- e1071::svm(tmp, factor(Y), probability = TRUE) 82 | prob.mat <- attr(stats::predict(model, t(mat), 83 | decision.values = FALSE, 84 | probability = TRUE), 85 | "probabilities") 86 | } 87 | if (!(classifier %in% c("svm", "rf"))) { 88 | stop("Classifier ", classifier, " is unknown.") 89 | } 90 | 91 | X <- c() 92 | Y <- c() 93 | 94 | XY_update <- lapply(seq_len(ncol(prob.mat)), function(j) { 95 | voteClass <- prob.mat[label==colnames(prob.mat)[j],] 96 | idx <- c() 97 | if (balance) { 98 | idx <- sample(seq_len(nrow(voteClass)), 99 | size=nrow(voteClass)*percent, replace = TRUE, 100 | prob=voteClass[,j]) 101 | } else { 102 | sampleSize <- round(stats::median(table(label))) 103 | if (nrow(voteClass) > sampleSize) { 104 | idx <- sample(seq_len(nrow(voteClass)), size=sampleSize* 105 | percent, replace = TRUE, 106 | prob=voteClass[,j]) 107 | } else { 108 | idx <- sample(seq_len(nrow(voteClass)), 109 | size=nrow(voteClass)*percent, 110 | replace = TRUE, prob=voteClass[,j]) 111 | } 112 | } 113 | 114 | list( 115 | X = mat[, rownames(voteClass)[idx]], 116 | Y = label[rownames(voteClass)[idx]] 117 | ) 118 | }) 119 | cur_X = lapply(XY_update, function(xy) { 120 | xy$X 121 | }) 122 | cur_Y = lapply(XY_update, function(xy) { 123 | xy$Y 124 | }) 125 | X <- do.call(cbind, cur_X) 126 | Y <- do.call(c, cur_Y) 127 | } 128 | 129 | model 130 | }) 131 | 132 | predictMat <- matrix(0, nrow=ncol(mat), ncol=length(table(label))) 133 | final <- c() 134 | predmat <- lapply(models, function(m) { 135 | if (classifier == "svm") { 136 | tmp <- attr(predict(m, 137 | newdata=t(mat), probability = TRUE), 138 | "prob")[,names(table(label))] 139 | } else { 140 | tmp <- predict(m, newdata=t(mat), 141 | type="prob")[,names(table(label))] 142 | } 143 | predictMat <<- predictMat + tmp 144 | }) 145 | 146 | if(prob==TRUE) { 147 | final <- apply(predictMat, 1, max) 148 | names(final) <- names(table(label))[apply(predictMat, 1, which.max)] 149 | } else { 150 | final <- names(table(label))[apply(predictMat, 1, which.max)] 151 | } 152 | 153 | return( 154 | list( 155 | final = final, 156 | models = models, 157 | prob = predictMat 158 | ) 159 | ) 160 | } 161 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | 2 | # scReClassify 3 | 4 | `scReClassify` is a post hoc cell type classification of single-cell 5 | RNA-sequencing data. Using semi-supervised learning algorithm, 6 | adaSampling, to correct cell type annotation from noise. 7 | 8 | # Getting started 9 | 10 | ## Vignette 11 | 12 | For more detailed instuction, you can find the vignette at our website: 13 | . 14 | 15 | ### Installation 16 | 17 | #### Bioconductor 18 | 19 | ```r 20 | if (!requireNamespace("BiocManager", quietly = TRUE)) 21 | install.packages("BiocManager") 22 | 23 | BiocManager::install("scReClassify") 24 | ``` 25 | 26 | #### GitHub 27 | 28 | ``` r 29 | devtools::install_github("SydneyBioX/scReClassify", build_opts = c("--no-resave-data", "--no-manual")) 30 | library(scReClassify) 31 | ``` 32 | 33 | For `devtool version 34 | < 2.0.0`, 35 | 36 | ``` r 37 | devtools::install_github("SydneyBioX/scReClassify", build_vignettes = TRUE) 38 | library(scReClassify) 39 | ``` 40 | 41 | 42 | ``` r 43 | suppressPackageStartupMessages({ 44 | library(scReClassify) 45 | library(SingleCellExperiment) 46 | library(SummarizedExperiment) 47 | }) 48 | ``` 49 | 50 | 51 | # Usage 52 | 53 | ![alt text](https://github.com/SydneyBioX/scReClassify/raw/master/img/scReClassify.jpg) 54 | 55 | 56 | Current version of this package is implemented to run with `svm` and `randomForest` classifiers. 57 | 58 | ## Load data 59 | 60 | ``` r 61 | data(gse87795_subset_sce) 62 | dat <- gse87795_subset_sce 63 | cellTypes <- gse87795_subset_sce$cellTypes 64 | 65 | # number of clusters 66 | nCs <- length(table(cellTypes)) 67 | 68 | # This demo dataset is already pre-processed 69 | dim(dat) 70 | ``` 71 | 72 | ## Part A. scReClassify (Demonstration) 73 | 74 | ### Dimension reduction 75 | 76 | ``` r 77 | reducedDim(dat, "matPCs") = matPCs(dat, assay = "logNorm", 0.7) 78 | ``` 79 | 80 | ### Synthetic noise (Demonstration purpose) 81 | 82 | Here in this example, we will synthetically generate varying degree of noise in sample labels. 83 | 84 | ``` r 85 | lab <- cellTypes 86 | 87 | set.seed(1) 88 | noisyCls <- function(dat, rho, cls.truth){ 89 | cls.noisy <- cls.truth 90 | names(cls.noisy) <- colnames(dat) 91 | for(i in seq_len(length(table(cls.noisy)))) { 92 | # class label starts from 0 93 | if (i != length(table(cls.noisy))) { 94 | cls.noisy[sample(which(cls.truth == names(table(cls.noisy))[i]), floor(sum(cls.truth == names(table(cls.noisy))[i]) * rho))] <- names(table(cls.noisy))[i+1] 95 | } else { 96 | cls.noisy[sample(which(cls.truth == names(table(cls.noisy))[i]), floor(sum(cls.truth == names(table(cls.noisy))[i]) * rho))] <- names(table(cls.noisy))[1] 97 | } 98 | } 99 | 100 | print(sum(cls.truth != cls.noisy)) 101 | return(cls.noisy) 102 | } 103 | 104 | cls.noisy01 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.1, lab) 105 | cls.noisy02 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.2, lab) 106 | cls.noisy03 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.3, lab) 107 | cls.noisy04 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.4, lab) 108 | cls.noisy05 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.5, lab) 109 | ``` 110 | 111 | ### Use scReClassify to correct mislabeled cell types. 112 | 113 | Here in this example, we will only use `Support Vector machine (svm)` as base classifier. 114 | 115 | #### Benchmark evaluation 116 | 117 | ``` r 118 | ################################### 119 | # SVM 120 | ################################### 121 | base <- "svm" 122 | set.seed(1) 123 | result = lapply(seq_len(10), function(j) { 124 | final <- multiAdaSampling(dat, cls.noisy01, reducedDimName = "matPCs", 125 | classifier=base, percent=1, L=10)$final 126 | ari01 <- mclust::adjustedRandIndex(lab, final) 127 | acc01 <- bAccuracy(lab, final) 128 | 129 | final <- multiAdaSampling(dat, cls.noisy02, reducedDimName = "matPCs", 130 | classifier=base, percent=1, L=10)$final 131 | ari02 <- mclust::adjustedRandIndex(lab, final) 132 | acc02 <- bAccuracy(lab, final) 133 | 134 | final <- multiAdaSampling(dat, cls.noisy03, reducedDimName = "matPCs", 135 | classifier=base, percent=1, L=10)$final 136 | ari03 <- mclust::adjustedRandIndex(lab, final) 137 | acc03 <- bAccuracy(lab, final) 138 | 139 | final <- multiAdaSampling(dat, cls.noisy04, reducedDimName = "matPCs", 140 | classifier=base, percent=1, L=10)$final 141 | ari04 <- mclust::adjustedRandIndex(lab, final) 142 | acc04 <- bAccuracy(lab, final) 143 | 144 | final <- multiAdaSampling(dat, cls.noisy05, reducedDimName = "matPCs", 145 | classifier=base, percent=1, L=10)$final 146 | ari05 <- mclust::adjustedRandIndex(lab, final) 147 | acc05 <- bAccuracy(lab, final) 148 | 149 | c( 150 | acc01 = acc01, 151 | acc02 = acc02, 152 | acc03 = acc03, 153 | acc04 = acc04, 154 | acc05 = acc05, 155 | ari01 = ari01, 156 | ari02 = ari02, 157 | ari03 = ari03, 158 | ari04 = ari04, 159 | ari05 = ari05 160 | ) 161 | }) 162 | 163 | result = do.call(rbind, result) 164 | acc = result[,seq_len(5)] 165 | colnames(acc) = seq(from=0.1,to=0.5,by=0.1) 166 | 167 | ari = result[,seq(from= 6, to = 10)] 168 | colnames(ari) = seq(from=0.1,to=0.5,by=0.1) 169 | 170 | 171 | plot.new() 172 | par(mfrow = c(1,2)) 173 | boxplot(acc, col="lightblue", main="SVM Accuracy", 174 | ylim=c(0.45, 1), xlab = "rho", ylab = "Accuracy") 175 | points(x=seq_len(5), y=c( 176 | bAccuracy(lab, cls.noisy01), 177 | bAccuracy(lab, cls.noisy02), 178 | bAccuracy(lab, cls.noisy03), 179 | bAccuracy(lab, cls.noisy04), 180 | bAccuracy(lab, cls.noisy05)), 181 | col="red3", pch=c(2,3,4,5,6), cex=1) 182 | boxplot(ari, col="lightblue", main="SVM ARI", 183 | ylim=c(0.25, 1), xlab = "rho", ylab = "ARI") 184 | points(x=seq_len(5), y=c( 185 | mclust::adjustedRandIndex(lab, cls.noisy01), 186 | mclust::adjustedRandIndex(lab, cls.noisy02), 187 | mclust::adjustedRandIndex(lab, cls.noisy03), 188 | mclust::adjustedRandIndex(lab, cls.noisy04), 189 | mclust::adjustedRandIndex(lab, cls.noisy05)), 190 | col="red3", pch=c(2,3,4,5,6), cex=1) 191 | ``` 192 | 193 | 194 | ## Part B. scReClassify (mislabeled cell type correction) 195 | 196 | ``` r 197 | # PCA procedure 198 | reducedDim(dat, "matPCs") = matPCs(dat, assay = "logNorm", 0.7) 199 | 200 | 201 | # run scReClassify 202 | set.seed(1) 203 | cellTypes.reclassify <- multiAdaSampling(dat, cellTypes, 204 | reducedDimName = "matPCs", 205 | classifier = "svm", percent = 1, L = 10) 206 | 207 | # Verification by marker genes 208 | End <- c("ITGA2B", "ITGB3") 209 | 210 | # check examples 211 | idx <- which(cellTypes.reclassify$final != cellTypes) 212 | 213 | cbind(original=cellTypes[idx], reclassify=cellTypes.reclassify$final[idx]) %>% 214 | DT::datatable() 215 | 216 | mat <- assay(dat, "logNorm") 217 | 218 | c1 <- mat[, which(cellTypes=="Endothelial Cell")] 219 | c2 <- mat[, which(cellTypes=="Erythrocyte")] 220 | c3 <- mat[, which(cellTypes=="Hepatoblast")] 221 | c4 <- mat[, which(cellTypes=="Macrophage")] 222 | c5 <- mat[, which(cellTypes=="Megakaryocyte")] 223 | c6 <- mat[, which(cellTypes=="Mesenchymal Cell")] 224 | cs <- rainbow(length(table(cellTypes))) 225 | 226 | 227 | # (example 1 E13.5_C14) 228 | ##### 229 | par(mfrow=c(1,2)) 230 | marker <- End[1] 231 | boxplot(c1[marker,], c2[marker,], c3[marker,], 232 | c4[marker,], c5[marker,], c6[marker,], 233 | col=cs, main=marker, 234 | names=c("Others", "Others", "Others", "Orignal", 235 | "Reclassified", "Others"), las=2, xlab = "Labels", 236 | ylab = "log2FPKM") 237 | points(5, mat[marker, which(colnames(mat) %in% "E13.5_C14")], 238 | pch=16, col="red", cex=2) 239 | 240 | marker <- End[2] 241 | boxplot(c1[marker,], c2[marker,], c3[marker,], 242 | c4[marker,], c5[marker,], c6[marker,], 243 | col=cs, main=marker, 244 | names=c("Others", "Others", "Others", "Orignal", 245 | "Reclassified", "Others"), las=2, xlab = "Labels", 246 | ylab = "log2FPKM") 247 | points(5, mat[marker, which(colnames(mat) %in% "E13.5_C14")], 248 | pch=16, col="red", cex=2) 249 | ``` 250 | 251 | 252 | # Reference 253 | scReClassify is published in BMC Genomics. Please refer to the following article for more details on method implementation and evaluation. 254 | 255 | Taiyun Kim, Kitty Lo, Thomas A. Geddes, Hani Jieun Kim, Jean Yee Hwa Yang & Pengyi Yang (2019) scReClassify: post hoc cell type classification of single-cell RNA-seq data. BMC Genomics, 20:913. https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-019-6305-x 256 | -------------------------------------------------------------------------------- /data/gse87795_subset_sce.rda: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/SydneyBioX/scReClassify/679abbb63ce3c141f2638b35a563e21b8a51ce4d/data/gse87795_subset_sce.rda -------------------------------------------------------------------------------- /img/scReClassify.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/SydneyBioX/scReClassify/679abbb63ce3c141f2638b35a563e21b8a51ce4d/img/scReClassify.jpg -------------------------------------------------------------------------------- /img/scReClassify_sticker.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/SydneyBioX/scReClassify/679abbb63ce3c141f2638b35a563e21b8a51ce4d/img/scReClassify_sticker.png -------------------------------------------------------------------------------- /inst/CITATION: -------------------------------------------------------------------------------- 1 | citHeader("To cite scReClassify in publications use:") 2 | citEntry(entry = "Article", 3 | title = "scReClassify: post hoc cell type classification of single-cell RNA-seq data", 4 | author = personList(as.person("Taiyun Kim"), 5 | as.person("Kitty Lo"), 6 | as.person("Thomas A Geddes"), 7 | as.person("Hani Jieun Kim"), 8 | as.person("Jean Yee Hwa Yang"), 9 | as.person("Pengyi Yang")), 10 | journal = "BMC Genomics", 11 | year = "2019", 12 | volume = "20", 13 | number = "913", 14 | url = "https://doi.org/10.1186/s12864-019-6305-x", 15 | textVersion = 16 | paste("Kim, T., Lo, K., Geddes, T. A., Kim, H. J., Yang, J. Y. H., Yang, P.", 17 | "scReClassify: post hoc cell type classification of single-cell RNA-seq data.", 18 | "BMC Genomics, 20, 913 (2019)", 19 | "doi: https://doi.org/10.1186/s12864-019-6305-x") 20 | ) -------------------------------------------------------------------------------- /inst/scReClassify.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/SydneyBioX/scReClassify/679abbb63ce3c141f2638b35a563e21b8a51ce4d/inst/scReClassify.jpg -------------------------------------------------------------------------------- /inst/scReClassify_sticker.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/SydneyBioX/scReClassify/679abbb63ce3c141f2638b35a563e21b8a51ce4d/inst/scReClassify_sticker.png -------------------------------------------------------------------------------- /man/bAccuracy.Rd: -------------------------------------------------------------------------------- 1 | % Generated by roxygen2: do not edit by hand 2 | % Please edit documentation in R/bAccuracy.R 3 | \name{bAccuracy} 4 | \alias{bAccuracy} 5 | \title{bAccuracy} 6 | \usage{ 7 | bAccuracy(cls.truth, final) 8 | } 9 | \arguments{ 10 | \item{cls.truth}{A character vector of true class label.} 11 | 12 | \item{final}{A vector of final classified label prediction from 13 | \code{multiAdaSampling}.} 14 | } 15 | \value{ 16 | An accuracy value. 17 | } 18 | \description{ 19 | This function calculates the accuracy of the prediction to the true label. 20 | } 21 | \examples{ 22 | data("gse87795_subset_sce") 23 | 24 | mat.expr <- gse87795_subset_sce 25 | cellTypes <- gse87795_subset_sce$cellTypes 26 | 27 | # Get dimension reduced matrix. We are using `logNorm` assay from `mat.expr`. 28 | mat.pc <- matPCs(mat.expr, assay = "logNorm") 29 | 30 | # Here we are using Support Vector Machine as a base classifier. 31 | result <- multiAdaSampling(mat.pc, cellTypes, classifier = "svm", 32 | percent = 1, L = 10) 33 | 34 | final <- result$final 35 | 36 | # Balanced accuracy 37 | bacc <- bAccuracy(cellTypes, final) 38 | 39 | } 40 | \author{ 41 | Pengyi Yang, Taiyun Kim 42 | } 43 | -------------------------------------------------------------------------------- /man/gse87795_subset_sce.Rd: -------------------------------------------------------------------------------- 1 | % Generated by roxygen2: do not edit by hand 2 | % Please edit documentation in R/data.R 3 | \docType{data} 4 | \name{gse87795_subset_sce} 5 | \alias{gse87795_subset_sce} 6 | \title{GSE827795 subset data} 7 | \format{ 8 | An object of class \code{SingleCellExperiment} with 1000 rows and 367 columns. 9 | } 10 | \usage{ 11 | gse87795_subset_sce 12 | } 13 | \description{ 14 | A SingleCellExperiment object containing a subset expression 15 | matrix of GSE827795. The data contains log2 transformed FPKM expression. 16 | 17 | GSE87795 is a mouse fetal liver development data containing 1000 genes, 18 | 367 cells and 6 cell types. 19 | 20 | The original GSE87795 data and the study details can be found at this 21 | \href{https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE87795}{link} 22 | } 23 | \keyword{datasets} 24 | -------------------------------------------------------------------------------- /man/matPCs.Rd: -------------------------------------------------------------------------------- 1 | % Generated by roxygen2: do not edit by hand 2 | % Please edit documentation in R/matPCs.R 3 | \name{matPCs} 4 | \alias{matPCs} 5 | \title{matPCs function} 6 | \usage{ 7 | matPCs(data, assay = NULL, percentVar = 0.8) 8 | } 9 | \arguments{ 10 | \item{data}{An expression matrix or a SingleCellExperiment object.} 11 | 12 | \item{assay}{An assay to select if \code{data} is a SingleCellExperiment 13 | object} 14 | 15 | \item{percentVar}{The percentage of variance threshold. This is used to 16 | select number of Principal Components.} 17 | } 18 | \value{ 19 | Dimensionally reduced matrix. 20 | } 21 | \description{ 22 | Performs PCA on a given matrix and returns a dimension reduced 23 | matrix which captures at least 80\% (default) of overall variability. 24 | } 25 | \details{ 26 | This function performs PCA to reduce the dimension of the gene 27 | expression matrix limited from 10 to 20 PCs. 28 | } 29 | \examples{ 30 | data("gse87795_subset_sce") 31 | 32 | mat.expr <- gse87795_subset_sce 33 | 34 | mat.pc <- matPCs(mat.expr, assay = "logNorm") 35 | 36 | # to capture at least 70\% of overall variability in the dataset, 37 | mat.dim.reduct.70 <- matPCs(mat.expr, assay = "logNorm", 0.7) 38 | 39 | } 40 | \author{ 41 | Pengyi Yang, Taiyun Kim 42 | } 43 | -------------------------------------------------------------------------------- /man/multiAdaSampling.Rd: -------------------------------------------------------------------------------- 1 | % Generated by roxygen2: do not edit by hand 2 | % Please edit documentation in R/multiAdaSampling.R 3 | \name{multiAdaSampling} 4 | \alias{multiAdaSampling} 5 | \title{multi Adaptive Sampling function} 6 | \usage{ 7 | multiAdaSampling( 8 | data, 9 | label, 10 | reducedDimName = NULL, 11 | classifier = "svm", 12 | percent = 1, 13 | L = 10, 14 | prob = FALSE, 15 | balance = TRUE, 16 | iter = 3 17 | ) 18 | } 19 | \arguments{ 20 | \item{data}{A dimension reduced matrix from \code{matPCs}.} 21 | 22 | \item{label}{A vector of label information for each sample.} 23 | 24 | \item{reducedDimName}{A name of the \code{reducedDim} to use. This must be 25 | specified if \code{data} is a SingleCellExperiment object.} 26 | 27 | \item{classifier}{Base classifier model, either "SVM" (\code{svm}) or "RF" 28 | \code{'rf'} is supported.} 29 | 30 | \item{percent}{Percentage of samples to select at each iteration.} 31 | 32 | \item{L}{Number of ensembles. Default to 10.} 33 | 34 | \item{prob}{logical flag to return sample's probabilities to each class.} 35 | 36 | \item{balance}{logical flag to if the cell types are balanced. 37 | If \code{FALSE}, down sample large cell types classes to the median of all class 38 | sizes.} 39 | 40 | \item{iter}{A number of iterations to perform adaSampling.} 41 | } 42 | \value{ 43 | A final prediction, probabilities for each cell type and the model 44 | are returned as a list. 45 | } 46 | \description{ 47 | Performs multiple adaptive sampling to train a classifier model. 48 | } 49 | \examples{ 50 | 51 | library(SingleCellExperiment) 52 | 53 | # Loading the data 54 | data("gse87795_subset_sce") 55 | 56 | mat.expr <- gse87795_subset_sce 57 | cellTypes <- gse87795_subset_sce$cellTypes 58 | 59 | # Get dimension reduced matrix. We are using `logNorm` assay from `mat.expr`. 60 | reducedDim(mat.expr, "matPCs") <- matPCs(mat.expr, assay = "logNorm") 61 | 62 | # Here we are using Support Vector Machine as a base classifier. 63 | result <- multiAdaSampling(mat.expr, cellTypes, reducedDimName = "matPCs", 64 | classifier = "svm", percent = 1, L = 10) 65 | } 66 | \author{ 67 | Pengyi Yang, Taiyun Kim 68 | } 69 | -------------------------------------------------------------------------------- /man/scReClassify.Rd: -------------------------------------------------------------------------------- 1 | % Generated by roxygen2: do not edit by hand 2 | % Please edit documentation in R/data.R 3 | \docType{package} 4 | \name{scReClassify} 5 | \alias{scReClassify} 6 | \title{scReClassify: a package for post hoc cell type classification of single-cell 7 | RNA-sequencing data.} 8 | \description{ 9 | A post hoc cell type classification tool to fine-tune cell type 10 | annotations generated by any cell type classification procedure with 11 | semi-supervised learning algorithm AdaSampling technique. 12 | 13 | The current version of scReClassify supports Support Vector Machine and 14 | Random Forest as a base classifier. 15 | } 16 | \seealso{ 17 | Useful links: 18 | \itemize{ 19 | \item Vignette available at: https://sydneybiox.github.io/scdney/ 20 | } 21 | } 22 | \author{ 23 | \strong{Maintainer}: 24 | \itemize{ 25 | \item Taiyun Kim (ORCID:0000-0002-5028-836X) 26 | \itemize{ 27 | \item Email: taiyun.kim91@gmail.com 28 | } 29 | } 30 | 31 | Authors: 32 | \itemize{ 33 | \item Pengyi Yang (ORCID: 0000-0003-1098-3138) 34 | } 35 | } 36 | -------------------------------------------------------------------------------- /tests/testthat.R: -------------------------------------------------------------------------------- 1 | library(testthat) 2 | library(scReClassify) 3 | 4 | test_check("scReClassify") 5 | -------------------------------------------------------------------------------- /tests/testthat/test-bAccuracy.R: -------------------------------------------------------------------------------- 1 | test_that( 2 | "Case bAccuracy.1: Missing parameters", 3 | { 4 | cls = 1:12 5 | label = 1:12 6 | 7 | expect_error(bAccuracy()) 8 | expect_error(bAccuracy(cls.truth = cls)) 9 | expect_error(bAccuracy(final = label)) 10 | } 11 | ) 12 | 13 | 14 | test_that( 15 | "Case bAccuracy.2: Different parameter lengths", 16 | { 17 | cls = 1:12 18 | label = 1:10 19 | 20 | expect_error(bAccuracy(cls, label)) 21 | } 22 | ) 23 | 24 | test_that( 25 | "Case bAccuracy.3: Expected outputs", 26 | { 27 | cls = 1:12 28 | label = 12:1 29 | exp_result1 = 0 30 | exp_result2 = 1 31 | 32 | expect_identical(exp_result1, bAccuracy(cls, label)) 33 | expect_identical(exp_result2, bAccuracy(cls, cls)) 34 | 35 | } 36 | ) 37 | -------------------------------------------------------------------------------- /tests/testthat/test-matPCs.R: -------------------------------------------------------------------------------- 1 | test_that( 2 | "Case matPCs.1: Missing parameters", 3 | { 4 | expect_error(matPCs()) 5 | expect_error(matPCs(percentVar = 0.8)) 6 | } 7 | ) 8 | 9 | test_that( 10 | "Case matPCs.2: Expected outputs", 11 | { 12 | set.seed(123) 13 | # 50 cells 400 genes 14 | mat = matrix(rnorm(20000), nrow = 50) 15 | pc = prcomp(mat) 16 | eigs = pc$sdev^2 17 | p = cumsum(eigs)/sum(eigs) 18 | 19 | result1 = min(which(p > 0.5)) 20 | result2 = min(which(p > 0.8)) 21 | # Since result2 > 20 22 | result2 = 20L 23 | 24 | # Number of PCs returned 25 | expect_identical(result1, ncol(matPCs(t(mat), percentVar = 0.5))) 26 | expect_identical(result2, ncol(matPCs(t(mat), percentVar = 0.8))) 27 | } 28 | ) 29 | -------------------------------------------------------------------------------- /tests/testthat/test-multiAdaSampling.R: -------------------------------------------------------------------------------- 1 | test_that( 2 | "Case multiAdaSampling.1: Missing parameters", 3 | { 4 | dat = matrix(rnorm(1200), nrow= 12) 5 | label = 1:ncol(dat) 6 | expect_error(multiAdaSampling()) 7 | expect_error(multiAdaSampling(data = dat)) 8 | expect_error(multiAdaSampling(label = label)) 9 | } 10 | ) 11 | 12 | test_that( 13 | "Case multiAdaSampling.2: Expected output", 14 | { 15 | set.seed(123) 16 | # very basic data with strong signal with 2 classes 17 | dat = matrix(rnorm(1200), ncol = 12) 18 | colnames(dat) = paste0("Cell", seq_len(12)) 19 | rownames(dat) = paste0("Gene", seq_len(100)) 20 | label = rep(c("CT_A", "CT_B"), each = 6) 21 | names(label) = colnames(dat) 22 | noise_label = label 23 | 24 | pc = matPCs(dat, percentVar = 0.8) 25 | result = multiAdaSampling( 26 | data = pc, 27 | label = noise_label, 28 | classifier = "svm" 29 | ) 30 | 31 | object_names = c("final", "models", "prob") 32 | prob_dim = c(12L, 2L) # 100 cells, 2 cell types 33 | 34 | expect_identical(object_names, names(result)) 35 | expect_identical(prob_dim, dim(result$prob)) 36 | expect_s3_class(result$models[[1]], "svm") 37 | expect_identical(12L, length(result$final)) 38 | } 39 | ) 40 | 41 | 42 | 43 | -------------------------------------------------------------------------------- /vignettes/scReClassify.Rmd: -------------------------------------------------------------------------------- 1 | --- 2 | title: "An introduction to scReClassify package" 3 | author: 4 | - name: Taiyun Kim 5 | affiliation: 6 | - School of Mathematics and Statistics, The University of Sydney 7 | - Computational Systems Biology Group, Children’s Medical Research Institute, Faculty of Medicine and Health, The University of Sydney 8 | - Charles Perkins Centre, The University of Sydney 9 | output: 10 | BiocStyle::html_document: 11 | toc_newpage: true 12 | fig_retina: NULL 13 | package: BiocStyle 14 | vignette: > 15 | %\VignetteIndexEntry{An introduction to scReClassify package} 16 | %\VignetteEngine{knitr::rmarkdown} 17 | %\VignetteEncoding{UTF-8} 18 | --- 19 | 20 | ```{r setup, include=FALSE} 21 | knitr::opts_chunk$set(echo = TRUE) 22 | ``` 23 | 24 | 25 | # Introduction 26 | 27 | 28 | 29 | `scReClassify` is a post hoc cell type classification of single-cell 30 | RNA-sequencing data to fine-tune cell type annotations generated by any cell 31 | type classification procedure. Typically, cell type identification relies on 32 | human inspection using combination of prior biological knowledge and 33 | computational techniques. Due to incompleteness of our current knowledge and the 34 | subjectivity involved in this process, a small amount of cells may be subject to 35 | mislabelling. Using semi-supervised learning algorithm, adaSampling, we are able 36 | to correct cell type annotations from various degree of noise. 37 | 38 | 39 | ![Overview of scReClassify methods](https://github.com/SydneyBioX/scReClassify/raw/master/img/scReClassify.jpg) 40 | 41 | # Installation 42 | 43 | Install the latest development version from GitHub using the `devtools` package: 44 | 45 | ```{r, eval = FALSE} 46 | if (!("devtools" %in% rownames(installed.packages()))) 47 | install.packages("devtools") 48 | 49 | library(devtools) 50 | devtools::install_github("SydneyBioX/scReClassify") 51 | ``` 52 | 53 | To install the Bioconductor version of scReClassify, enter the following to your 54 | R console. 55 | 56 | ```{r, eval = FALSE} 57 | if (!requireNamespace("BiocManager", quietly = TRUE)) 58 | install.packages("BiocManager") 59 | BiocManager::install("scReClassify") 60 | ``` 61 | 62 | # Loading packages and data 63 | 64 | ```{r} 65 | suppressPackageStartupMessages({ 66 | library(scReClassify) 67 | library(DT) 68 | library(mclust) 69 | library(dplyr) 70 | library(SummarizedExperiment) 71 | library(SingleCellExperiment) 72 | }) 73 | 74 | data("gse87795_subset_sce") 75 | 76 | dat <- gse87795_subset_sce 77 | cellTypes <- gse87795_subset_sce$cellTypes 78 | ``` 79 | 80 | `gse87795_subset_sce` is a `SingleCellExperiment` object of a mouse fetal liver 81 | development data deposited at Gene Expression Omnibus respository with accession 82 | ID GSE87795. The cell type information can be found on the `colData` of the 83 | `SingleCellExperiment` object. 84 | 85 | 86 | ```{r} 87 | # Cell types 88 | table(cellTypes) 89 | 90 | # We set the number of clusters 91 | nCs <- length(table(cellTypes)) 92 | nCs 93 | 94 | # This demo dataset is already pre-processed 95 | dim(dat) 96 | ``` 97 | 98 | There are `r nCs` cell types, `r ncol(dat)` cells and `r nrow(dat)` number of 99 | genes. 100 | 101 | 102 | # Part A. scReClassify (Demonstration with synthetic mislabels) 103 | 104 | ## Dimension reduction 105 | 106 | Prior to running scReClassify, we perform dimension reduction. `matPCs` is a 107 | tool in scReClassify to simplify this process. In this function, a dimension 108 | reduced matrix is returned with `n` principal components (PCs), where `n` is the 109 | number of principal components (PCs) that by sum explains at least 70% variance. 110 | 111 | The function accepts either a `matrix` or a `SingleCellExperiment` object. If 112 | the `data` parameter is a `SingleCellExperiment` object, an `assay` variable 113 | must be specified to perform dimension reduction on the correct assay. If the 114 | `SingleCellExperiment` object `data` already has a 'PCA' in `reducedDimNames()`, 115 | the 'PCA' matrix of `n` columns are returned. 116 | 117 | 118 | ```{r} 119 | reducedDim(dat, "matPCs") <- matPCs(dat, assay = "logNorm", 0.7) 120 | ``` 121 | 122 | ## Synthetic noise (Demonstration purpose) 123 | 124 | Here in this example, we will synthetically generate varying degree of 125 | noise (10-50%) in sample labels. The purpose here is to simulate different level 126 | of mislabeling in the data. Given a cell type label `cls.truth`, `noisyCls` 127 | function will randomly select a `rho` percentage of cells from a given cell type 128 | and relabel to other cell types. 129 | 130 | Here, we create different degree of noise from 10% to 50%. 131 | 132 | ```{r} 133 | lab <- cellTypes 134 | 135 | set.seed(1) 136 | # Function to create noise in the cell type label 137 | noisyCls <- function(dat, rho, cls.truth){ 138 | cls.noisy <- cls.truth 139 | names(cls.noisy) <- colnames(dat) 140 | 141 | for(i in seq_len(length(table(cls.noisy)))) { 142 | # class label starts from 0 143 | if (i != length(table(cls.noisy))) { 144 | cls.noisy[sample(which(cls.truth == names(table(cls.noisy))[i]), 145 | floor(sum(cls.truth == names(table(cls.noisy))[i])* 146 | rho))] <- names(table(cls.noisy))[i+1] 147 | } else { 148 | cls.noisy[sample(which(cls.truth == names(table(cls.noisy))[i]), 149 | floor(sum(cls.truth == names(table(cls.noisy))[i])* 150 | rho))] <- names(table(cls.noisy))[1] 151 | } 152 | } 153 | 154 | print(sum(cls.truth != cls.noisy)) 155 | return(cls.noisy) 156 | } 157 | 158 | cls.noisy01 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.1, lab) 159 | cls.noisy02 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.2, lab) 160 | cls.noisy03 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.3, lab) 161 | cls.noisy04 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.4, lab) 162 | cls.noisy05 <- noisyCls(t(reducedDim(dat, "matPCs")), rho=0.5, lab) 163 | ``` 164 | 165 | With `noisyCls` function, we have relabeled `r cls.noisy01`, `r cls.noisy01`, 166 | `r cls.noisy01`, `r cls.noisy01` and `r cls.noisy01` number of cells for `rho` 167 | equal to 0.1, 0.2, 0.3, 0.4 and 0.5 respectively. 168 | 169 | 170 | ## Use scReClassify to correct mislabeled cell types. 171 | 172 | Here in this example, we will only use `Support Vector machine (svm)` as 173 | base classifier. 174 | 175 | #### Benchmark evaluation 176 | 177 | To benchmark scReClassify, we perform scReclassify to all degree of noise with 178 | 10 repeats. We measure the accuracy of scReClassify and the Adjusted Rand Index 179 | (ARI) to measure the concordance of the reclassified cell type to the true cell 180 | type label. 181 | 182 | ```{r} 183 | ################################### 184 | # SVM 185 | ################################### 186 | base <- "svm" 187 | set.seed(1) 188 | result = lapply(seq_len(10), function(j) { 189 | final <- multiAdaSampling(dat, cls.noisy01, reducedDimName = "matPCs", 190 | classifier=base, percent=1, L=10)$final 191 | ari01 <- mclust::adjustedRandIndex(lab, final) 192 | acc01 <- bAccuracy(lab, final) 193 | 194 | final <- multiAdaSampling(dat, cls.noisy02, reducedDimName = "matPCs", 195 | classifier=base, percent=1, L=10)$final 196 | ari02 <- mclust::adjustedRandIndex(lab, final) 197 | acc02 <- bAccuracy(lab, final) 198 | 199 | final <- multiAdaSampling(dat, cls.noisy03, reducedDimName = "matPCs", 200 | classifier=base, percent=1, L=10)$final 201 | ari03 <- mclust::adjustedRandIndex(lab, final) 202 | acc03 <- bAccuracy(lab, final) 203 | 204 | final <- multiAdaSampling(dat, cls.noisy04, reducedDimName = "matPCs", 205 | classifier=base, percent=1, L=10)$final 206 | ari04 <- mclust::adjustedRandIndex(lab, final) 207 | acc04 <- bAccuracy(lab, final) 208 | 209 | final <- multiAdaSampling(dat, cls.noisy05, reducedDimName = "matPCs", 210 | classifier=base, percent=1, L=10)$final 211 | ari05 <- mclust::adjustedRandIndex(lab, final) 212 | acc05 <- bAccuracy(lab, final) 213 | 214 | c( 215 | acc01 = acc01, 216 | acc02 = acc02, 217 | acc03 = acc03, 218 | acc04 = acc04, 219 | acc05 = acc05, 220 | ari01 = ari01, 221 | ari02 = ari02, 222 | ari03 = ari03, 223 | ari04 = ari04, 224 | ari05 = ari05 225 | ) 226 | }) 227 | 228 | result = do.call(rbind, result) 229 | acc = result[,seq_len(5)] 230 | colnames(acc) = seq(from=0.1,to=0.5,by=0.1) 231 | 232 | ari = result[,seq(from= 6, to = 10)] 233 | colnames(ari) = seq(from=0.1,to=0.5,by=0.1) 234 | 235 | ``` 236 | 237 | 238 | We can visualise the performance of the scReClassify. The boxes represent the 239 | accuracy and the ARI after scReClassify. The red markers indicate the baseline 240 | (prior to scReClassify). 241 | 242 | 243 | ```{r} 244 | plot.new() 245 | par(mfrow = c(1,2)) 246 | boxplot(acc, col="lightblue", main="SVM Accuracy", 247 | ylim=c(0.45, 1), xlab = "rho", ylab = "Accuracy") 248 | points(x=seq_len(5), y=c( 249 | bAccuracy(lab, cls.noisy01), 250 | bAccuracy(lab, cls.noisy02), 251 | bAccuracy(lab, cls.noisy03), 252 | bAccuracy(lab, cls.noisy04), 253 | bAccuracy(lab, cls.noisy05)), 254 | col="red3", pch=c(2,3,4,5,6), cex=1) 255 | boxplot(ari, col="lightblue", main="SVM ARI", 256 | ylim=c(0.25, 1), xlab = "rho", ylab = "ARI") 257 | points(x=seq_len(5), y=c( 258 | mclust::adjustedRandIndex(lab, cls.noisy01), 259 | mclust::adjustedRandIndex(lab, cls.noisy02), 260 | mclust::adjustedRandIndex(lab, cls.noisy03), 261 | mclust::adjustedRandIndex(lab, cls.noisy04), 262 | mclust::adjustedRandIndex(lab, cls.noisy05)), 263 | col="red3", pch=c(2,3,4,5,6), cex=1) 264 | ``` 265 | 266 | The plot shows that with scReClassify, cell type information have been refined 267 | (boxes are higher than the red markers). The scReClassified results show 268 | higher accuracy across noise levels 0.1 - 0.4 (i.e. closer to the true label). 269 | With the noise level 0.5, it is showing similar accuracy which is as expected 270 | because the initial label contains equal amount of true and false information 271 | and thus making it difficult for the algorithm to learn the true label. This 272 | shows that scReClassify is also robust to noisy cell type labels. 273 | 274 | 275 | # Part B. scReClassify (mislabeled cell type correction) 276 | 277 | scReClassify has shown promising result with the synthetic noise we have 278 | created. Here we will use scReClassify on the actual cell type label from public 279 | repository. The data we will use is a mouse fetal liver dataset from GEO with an 280 | accession ID GSE87795. 281 | 282 | 283 | ```{r} 284 | # PCA procedure 285 | reducedDim(dat, "matPCs") <- matPCs(dat, assay = "logNorm", 0.7) 286 | 287 | 288 | # run scReClassify 289 | set.seed(1) 290 | cellTypes.reclassify <- multiAdaSampling(dat, cellTypes, 291 | reducedDimName = "matPCs", 292 | classifier = "svm", percent = 1, L = 10) 293 | 294 | # Verification by marker genes 295 | End <- c("ITGA2B", "ITGB3") 296 | ``` 297 | 298 | Below is a table of cell type labels classified to a different cell types after 299 | scReClassify. 300 | 301 | ```{r} 302 | # check examples 303 | idx <- which(cellTypes.reclassify$final != cellTypes) 304 | 305 | cbind(original=cellTypes[idx], reclassify=cellTypes.reclassify$final[idx]) %>% 306 | DT::datatable() 307 | ``` 308 | 309 | 310 | Here, we visualise the expression level of the a cells that is reclassified for 311 | demontration purpose. The box plots are the marker gene expression levels 312 | grouped by cells types. The expression level of the reclassified cell (Cell ID: 313 | E13.5_C14) are highlighted as red marker. 314 | 315 | ```{r} 316 | mat <- assay(dat, "logNorm") 317 | 318 | c1 <- mat[, which(cellTypes=="Endothelial Cell")] 319 | c2 <- mat[, which(cellTypes=="Erythrocyte")] 320 | c3 <- mat[, which(cellTypes=="Hepatoblast")] 321 | c4 <- mat[, which(cellTypes=="Macrophage")] 322 | c5 <- mat[, which(cellTypes=="Megakaryocyte")] 323 | c6 <- mat[, which(cellTypes=="Mesenchymal Cell")] 324 | cs <- rainbow(length(table(cellTypes))) 325 | 326 | 327 | # (example 1 E13.5_C14) 328 | ##### 329 | par(mfrow=c(1,2)) 330 | marker <- End[1] 331 | boxplot(c1[marker,], c2[marker,], c3[marker,], 332 | c4[marker,], c5[marker,], c6[marker,], 333 | col=cs, main=marker, 334 | names=c("Others", "Others", "Others", "Orignal", 335 | "Reclassified", "Others"), las=2, xlab = "Labels", 336 | ylab = "log2FPKM") 337 | points(5, mat[marker, which(colnames(mat) %in% "E13.5_C14")], 338 | pch=16, col="red", cex=2) 339 | 340 | marker <- End[2] 341 | boxplot(c1[marker,], c2[marker,], c3[marker,], 342 | c4[marker,], c5[marker,], c6[marker,], 343 | col=cs, main=marker, 344 | names=c("Others", "Others", "Others", "Orignal", 345 | "Reclassified", "Others"), las=2, xlab = "Labels", 346 | ylab = "log2FPKM") 347 | points(5, mat[marker, which(colnames(mat) %in% "E13.5_C14")], 348 | pch=16, col="red", cex=2) 349 | ``` 350 | As shown in the boxplots above, the expression level of the reclassified cell 351 | (red dot) is similar to the expression levels of the reclassified cell types in 352 | a marker gene. This highlights that the E13.5_C14 cell has a similar expression 353 | profiles to the reclassified cell types rather than its originally labeled cell 354 | type. Thus, we were able to identify that E13.5_C14 potentially belongs to the 355 | reclassified cell type with scReClassify. 356 | 357 | 358 | # SessionInfo 359 | 360 | ```{r} 361 | sessionInfo() 362 | ``` 363 | 364 | 365 | --------------------------------------------------------------------------------