Younis Lab · CMU-Q
Spring 2026 to now
Dissecting the dark matter of splicing
One gene, many messages, alternative splicing is how a single DNA sequence encodes a whole repertoire of proteins. I analyse differential splicing across cell states with four independent algorithms, leafcutter, DEXSeq, MAJIQ and rMATs, four dialects of the same biological story. My current project builds a confidence matrix, an XGBoost model trained on all four tools' outputs that scores every candidate splicing change and surfaces the biologically relevant ones as a single, exhaustive catalogue for the wet lab to chase.
What I do in the lab
- Consensus calls, run all four splicing algorithms on the same cell-state comparisons and reconcile where they disagree, so a change is only called when the tools line up.
- Confidence matrix with XGBoost, a gradient-boosted model that learns from the four tools' outputs, junction reads, exon-level counts, PSI estimates, to score how confident we are in every splicing change and rank them for the wet lab to chase.
- Dry lab → wet lab, the breast-cancer splicing analysis starts as dry lab in Nepal, and every candidate finding gets validated at the bench in Doha. Same story, both sides of the pipette.
- Lab tooling, Python and R tools for the splicing pipeline: reproducible analysis scripts, figure pipelines and data prep the whole group actually uses.
