Workflows for MorPhiC bulk RNAseq data analysis
Explore data and generate meta data
Bulk RNA-seq DE testing by DESeq2
DE testing was conducted separately for different model systems (e.g., Cortical brain or Extra-embryonic primitive syncytial cells), oxygen conditions (normoxia or hypoxia), or time points.
Given a particular model system, oxygen condition, and time point, DESeq2 was run on the level of DE group. A DE group typically includes the samples belonging to one batch (i.e., one run id). Though if wild type samples from different batches are more similar to each other than the knock out samples (judged by PC plots), we combine samples of these batches as one DE group, in order to borrow information across batches.
For datasets produced by JAX, the major type of DE testing is between knockout samples and wild type samples, and the second type of DE testing is between normoxia and hypoxia conditions.
For ISL1_OE dataset produced by MSK, the major type of DE testing is between ISL1 OE samples and control samples, and the second type of DE testing is between knockout samples and wild type samples among the control samples.
For the dataset produced by NW, the major type of DE testing is between samples and control samples, and the second type of DE testing is between two different ways of doing the knockout.
For the dataset produced by UCSF, the DE testing is between each two consecutive time points.
Certain samples are excluded from the DE analysis on JAX and USCF datasets. On JAX datasets, these are either wild type samples with fewer than 2 matching KO samples or KO samples with fewer than 2 matching wild type samples. On the dataset produced by UCSF, these are samples not part of the EB time course.
To make the results more reliable, genes with low expression are excluded from DE testing or have adjusted pvalue marked as NA.
There are multiple steps involved in the filtering procedure:
-
For each gene, compute the 75% quantile of its expression among samples under each model system. If a gene has 75% quantile being 0 under all model systems in the dataset, exclude the gene from further analysis.
-
Before running DESeq2, based on the samples in the DE group, there are two steps excluding certain genes from DESeq2:
-
Genes with 75% quantile of expression being 0.
-
Genes with nonzero expression in fewer than 4 samples.
-
-
DESeq2 internally excludes genes with mean normalized counts below a filtering threshold from multiple testing adjustment, and these genes will have adjusted pvalue being NA.
-
After running DESeq2, for the results from a specific contrast (e.g., KO of one gene vs. WT), if one gene has zero expression in at least 4 samples, its adjusted pvalue is assigned as NA.
Visulization of the DE results, in two aspects:
-
Scatterplots showing Knockout effect on the knockout genes themselves, and the number of DE genes
-
Volcano plots showing DE genes.
Functional category enrichment analysis