Index
73 | 74 |
75 |
76 |
77 |
78 |
79 |
70 |
71 |
72 |
80 |
52 | 
53 | 
54 |
2 |
3 | # Genes2Genes
4 |
5 | ## A new framework for aligning single-cell trajectories of gene expression
6 | G2G aims to guide downstream comparative analysis of single-cell reference and query systems along any axis of progression (e.g. pseudotime).
7 | This is done by employing a new dynamic programming (DP) based alignment algorithm which unifies dynamic time warping (DTW) and gap modelling to capture both matches and mismatches between time points. Our DP algorithm incorporates a Bayesian information-theoretic scoring scheme with a five-state probabilistic machine to generate an optimal sequential alignment between a reference trajectory (R) and query trajectory (Q) of a given gene in terms of their scRNA-seq expression. In this way, G2G framework infers a fully-descriptive alignment for each gene of a specified gene set, gene clusters of different alignment patterns, an average (cell-level) alignment across all gene alignments, and further statistics to support downstream analysis (e.g. ranking of genes based on their alignment similarities).
8 |
9 | 
10 |
11 | G2G framework can perform comparisons of gene expression dynamics across pseudotime such as:
12 |
52 |
53 |
54 |
' + 106 | '' + 107 | _("Hide Search Matches") + 108 | "
" 109 | ) 110 | ); 111 | }, 112 | 113 | /** 114 | * helper function to hide the search marks again 115 | */ 116 | hideSearchWords: () => { 117 | document 118 | .querySelectorAll("#searchbox .highlight-link") 119 | .forEach((el) => el.remove()); 120 | document 121 | .querySelectorAll("span.highlighted") 122 | .forEach((el) => el.classList.remove("highlighted")); 123 | localStorage.removeItem("sphinx_highlight_terms") 124 | }, 125 | 126 | initEscapeListener: () => { 127 | // only install a listener if it is really needed 128 | if (!DOCUMENTATION_OPTIONS.ENABLE_SEARCH_SHORTCUTS) return; 129 | 130 | document.addEventListener("keydown", (event) => { 131 | // bail for input elements 132 | if (BLACKLISTED_KEY_CONTROL_ELEMENTS.has(document.activeElement.tagName)) return; 133 | // bail with special keys 134 | if (event.shiftKey || event.altKey || event.ctrlKey || event.metaKey) return; 135 | if (DOCUMENTATION_OPTIONS.ENABLE_SEARCH_SHORTCUTS && (event.key === "Escape")) { 136 | SphinxHighlight.hideSearchWords(); 137 | event.preventDefault(); 138 | } 139 | }); 140 | }, 141 | }; 142 | 143 | _ready(SphinxHighlight.highlightSearchWords); 144 | _ready(SphinxHighlight.initEscapeListener); 145 | -------------------------------------------------------------------------------- /doc/build/html/genindex.html: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 |A framework for single-cell pseudotime trajectory alignment
76 | 91 |Genes2Genes (G2G) is a new Python framework for aligning single-cell pseudotime trajectories of gene expression between any reference and query for a pairwise comparison such as:
92 |93 |99 |94 |
98 |- 95 |
Organoid vs. Reference tissue
- 96 |
Control vs. Treatment
- 97 |
Healthy vs. Disease
A single-cell trajectory describes the transcriptomic state of cells along some axis of progression (such as time), due to undergoing some dynamic process (e.g. cell differentiation, treatment response, or disease infection). Given an scRNA-seq profile, there are various tools available today to infer such trajectory by estimating a pseudo ordering of the cells along an axis, commonly referred to as ‘pseudotime’. The pseudotime axis of a trajectory can be descritized to represent it as a sequence of discrete time points. Given two such discrete pseudotime sequences of two trajectories, a pairwise alignment between them defines a non-linear mapping between their time points. This mapping could have 1-to-1 matches as well as 1-to-many/many-to-1 matches (a.k.a warps) between the time points, while unmapping the time points which have significantly different transcriptomic states. Below is an example visualization of two cell differentiation trajectories.
102 |
103 | For two trajectories representing single lineages as above, G2G generates an optimal pairwise trajectory alignment that captures the matches and mismatches between their time points in sequential order, allowing a user to quantify the degree of similarity between them.
104 |
105 | G2G defines 5 different states of alignment between any two R and Q time points, corresponding to all possible match and mismatch states. They are: 1-to-1 match (M), 1-to-many match (V), many-to-1 match (W), insertion (I) and deletion (D). Here, I or D refer to a mismatched time point in Q or R, respectively. These states jointly cover the alignment states defined in classical dynamic time warping and biological sequence alignment.
110 | Accordingly, we can describe any trajectory alignment as a 5-state alignment string. For example, the 5-state alignment string of the above illustrated trajectory alignment is:
114 |IIIMMMWWWIIIDDDMMMIIIIDDDD
115 | This G2G alignment string enables us to identify the time regions of match and mismatch along the trajectories. For instance, we can interpret the above illustrated alignment as follow – R and Q trajectories have mid and late mismatches, with the early stage of Q being mismatched, yet starting to match to the early stage of R at the middle of Q’s trajectory. Overall, there are 9 R and Q pseudotime pairs getting matched (with 34.62% alignment similarity).
118 |Given an scRNA-seq dataset with their pseudotime estimates and a specified set of genes (e.g. all transcription factors, highly variable genes, biological/signaling pathway genes), G2G generates fully-descriptive alignments for each gene (i.e. gene-level alignment), as well as an average (aggregate) alignment (i.e. cell-level alignment) across all genes.
122 |Below is an example gene-level alignment of the gene JUNB in T cell differentiation between a pan-fetal reference and an artificial thymic organoid system:
123 |
124 | IIIIIDMMMMMMMMIDDDDD
128 | Each gene-level alignment carries its 5-state string, an alignment similarity percentage statistic, and the optimal alignment cost (in nits – the unit measure of information). For the above gene, the aligment similarity is 40%, and the total cost of alignment is 53.47 nits. When the degree of difference in gene expression between the reference and query is high, the alignment cost will also be high.
131 |G2G uses the inferred gene-level alignments to inform:
132 |The degree of similarity between the two profiles as an average percentage of alignment similarity across all the genes tested,
An aggregate cell-level alignment across all genes to inform the average states of match and mismatch between the two profiles (again represented by a 5-state string),
A ranked list of genes across time (from the most distant to most similar) based on their alignment similarity percentage statistic,
The diversity of different alignment patterns in genes, by clustering gene-level alignments to identify different matching and mismatching patterns along time,
between the two single-cell reference and query profiles in comparison.
139 |For further downstream analysis, G2G provides a wrapper function to check gene-set overrepresentation analysis of the identified gene-clusters and the list of the top distant (differentially-expressed) genes across time, using GSEApy Enrichr interface. The user is also able to compute an average alignment across any gene subset of their interest.
140 |Note: G2G has been developed only for single-lineage trajectory comparison. In the case of a trajectory with multiple branches, we recommend separating out the singe-lineage branches before any pairwise comparison.
141 |We employ a dynamic programming (DP) based algorithm that can capture both matches and mismatches in gene expression in a unified way. This combines the classical Gotoh’s algorithm for biological sequence alignment with dynamic time warping. Our DP algorithm uses a Bayesian information-theoretic scoring scheme based on the minimum message length criterion to generate an optimal alignment between two gene trajectories. This scheme evaluates the distributional similarity of gene expression between R and Q for each pair of time points, in terms of both their mean and variance of expression modelled as Gaussian distributions. 145 | For more details on the methods, please see our manuscript.
146 |For now, G2G needs to be installed from GitHub:
151 |pip install git+https://github.com/Teichlab/Genes2Genes.git
152 | The package will be made available soon on PyPi.
155 |Input to Gene2Genes
156 |G2G takes reference and query input data as anndata objects, where each adata object has:
log1p normalised gene expression stored at adata.X
pseudotime estimates of the cells stored as adata.obs['time']
The user can estimate pseudotime of the cells in their datasets using any suitable method available (such as Diffusion pseudotime, Palantir, GPLVM, Monocle etc.). 162 | For better visualisation and interpretation of the alignment results, we recommend the data to be annotated with their cell types (manually and/or using an automatic annotation tool such as CellTypist).
163 |Please refer to our Tutorial for an example analysis between a reference and query dataset from literature.
164 |Our manuscript is currently available as a preprint at bioRxiv:
168 |Sumanaweera, D., Suo, C., Cujba, A.M., Muraro, D., Dann, E., Polanski, K., Steemers, A.S., Lee, W., Oliver, A.J., Park, J.E. and Meyer, K.B., 2023. Gene-level alignment of single cell trajectories. bioRxiv, pp.2023-03.
169 |This publication is part of the Human Cell Atlas
170 |Marie Skłodowska-Curie grant agreement No: 101026506 (Marie Curie Individual Fellowship) under the European Union’s Horizon 2020 research and innovation programme; Wellcome Trust Ph.D. Fellowship for Clinicians; Wellcome Trust (WT206194); ERC Consolidator Grant (646794); Wellcome Sanger Institute’s Translation Committee Fund.
174 |