I study how cells decide where to end a messenger RNA, and I build the computational tools to read those decisions at scale.
Most genes can end their messenger RNA in more than one place. This is called alternative polyadenylation, and the choice is not cosmetic: it changes how long the 3' untranslated region is, which regulatory sites the transcript carries, and ultimately how much protein the cell makes. Get it wrong and you can tip a cell toward cancer, neurodegeneration, or heart failure.
My work makes those choices visible. The diagram below shows the core idea in a single switch. Below that is PolyAMiner-Bulk, the deep-learning method I built to detect these shifts across the whole genome from ordinary RNA-seq data.
Most recent. My 2026 ARVO work applies this same lens to corneal wound repair: alternative polyadenylation, now literally in the eye.
Interactive
Toggle which polyadenylation site the cell uses and watch what happens to the transcript and its protein output.
Method
A deep-learning method that decodes alternative polyadenylation dynamics from ordinary bulk RNA-seq, no specialized library prep required.
Calls 3'UTR shortening and lengthening across every expressed gene, not just a curated panel.
A learned model separates true polyadenylation shifts from coverage noise, cutting false positives.
Per-gene APA indices and differential results you can take straight into a figure.
Published step by step in STAR Protocols so any lab can run it.
Where it matters
I am always glad to talk RNA, imaging, or clinical AI.