Model based clustering of RNA-seq data
GTEX V6 analysis
The Genotype Tissue Expression (GTEx) Project is a large scale project collecting tissue samples from actual human tissues. Check the paper.
We apply model based clustering of the bulk-RNA GTEx V6 data (check the official release: GTEx portal). The data contained 8555 tissue samples coming from 53 different tissues and we focussed on 16069 genes chosen using a filtering criterion. Below, we present the cluster analysis and Structure plot visualizations for number of clusters K=15 on all the tissues as well as clustering of brain samples using K=4.
An alternative representation of the clustering besides the Structure plot is t-SNE representation. We present the results from applying t-SNE on the original GTEx read counts data and also on the topic proportions from the topic model fit.
We annotate the genes that drive the clusters. For each cluster, we find a set of few top genes that distinguish that cluster from the rest (we term these as cluster annotating genes).
Jaitin et al (2014) single cell analysis
We apply the model based clustering on the dataset due to Jaitin et al 2014, paper.
Jaitin et al sequenced over 4000 single cells from mouse spleen. Following the original authors protocol, we also filtered out 16 genes that they found to show significant batch-specific expression. Here we analyze 1041 of these cells that were categorized as CD11c+ in the sorting markers column of their data (link), and which had total number of reads mapping to non-ERCC genes greater than 600. (We believe these cells correspond roughly to the 1040 cells in their Figure S7.)
Below we fit our model on the above data for K=7.
Deng et al (2014) single cell analysis
We apply the model based clustering on the dataset due to Deng et al 2014, paper.
Deng et al collected expression data from individual cells from zygote to blastocyst stages of mouse preimplantation development. Deng et al’s analysis focussed particularly on allele-specific expression from the two contributing mouse strains (CAST/EiJ and C57BL/6J).
Here we present the topic model fit and Structure plot on this data for a range of values of K (number of clusters) to see how patterns change with increasing number of topics.
The cluster annotations of the clusters for the data are given in the script below.
Other applications of CountClust
We apply a batch correction procedure
BatchCorrectedCounts() to remove known technical effects using a voom type framework in the package
CountClust. We present 3 simulation scenarios to present the effectiveness of the batch correction mechanism.
We also validated the clusters we obtained in GTEx V6 and the Deng 2014 data by considering some of the genes with interesting biological properties and checking the trends of log of expression across the different tissue samples for that gene.