Translational
Bioinformatics Lab

Tumors grown outside the patient as organoids and cell lines, characterized against the tissue they came from, tested with drugs and radiation, then biobanked and shared.

In rectal cancer, paired biopsies, irradiated organoids and public cohorts show where the response signal lives: less in baseline expression than assumed, more in the change under treatment.

Gut and oral microbiome, blood transcriptome and plasma cell-free DNA from the same untreated patients, to learn whether a liquid biopsy reports the tumor or the host carrying it.

Cross-attention models that fuse mutations, expression, protein and histology to predict drug sensitivity, and the audit that goes with them: do the labels mean what we assume, and where does a linear baseline win?

The same genomics where the biology is interesting: carbonate-precipitating bacteria in livestock-waste metagenomes, and plant root transcriptomes under a polyphenol.

What a multi-omics model actually learns when it predicts drug sensitivity across cancers: lineage and differentiation programs rather than driver mutations, a shared axis that carries over to cancer types it never saw, and a stemness index that tracks sensitivity in measured cell-line pharmacology.