Omics Distillery
The Concept
I have an some RNA-seq data from an in vitro drug candidate screening experiment. I want to know:
-
why is a candidate working?
-
what is the mechanism of action?
-
will it be working in patients?
-
are there subgroups of patients that will respond positively to it?
The conventional way
-
Run data processing pipelines to turn sequencing reads into counts.
-
Differential gene expression analysis, make volcano plot and heatmap.
-
Pathway & gene ontology enrichment analysis, top pathways have 10 genes and 5 are common signal transduction genes.
-
Find an interesting gene set and do Gene Set Enrichment Analysis.
-
Maybe transcription factor enrichment analysis.
-
Maybe network analysis, if there's enough samples.
-
Read papers and hope that some of the differentially expressed genes may connect to a potential mechanism.
The futuristic way
-
Run data processing pipelines to turn sequencing reads into counts.
-
Use transfer learning to learn representation of expression profile and infer functional changes based on similarity with previous observations.
-
Use generative models to deconvolute cell types and cell type-specific gene regulatory network changing between samples.
-
Multiple specialised models to identify metabolic states, surface protein interactome, signalling networks.
-
Models to extrapolate transcriptome to metabolome, proteome, epigenome, phenome etc. to generate hypothesis for the next experiment.