I study which tumor cells survive therapy, using public single-cell and spatial data. I test my tools before I trust their results, and I release everything needed to check my work.
Website · Google Scholar · ORCID
| Project | Question and result | Output |
|---|---|---|
| dmg_atlas | Which diffuse midline glioma cell states replicate across 6 studies and 311,618 cells? Only cycling and hypoxia met a pre-registered replication rule. | [Preprint] · Zenodo |
| sc-cnv-fps | How often do copy-number callers label normal cells as malignant? I ran three callers on 14 tumor-free datasets; an empirical null cut false calls about 90-fold. | [Preprint] |
| dmg_mh_states | Do mouse DMG models reproduce human tumor cell states? | [Status] |
| ews_atlas | A harmonized single-cell atlas of Ewing sarcoma. | [Status] |
| cfm_eval | Do single-cell foundation models work zero-shot on cancer data? | [Status] |
| fezf2-loss-multi-omics | How does Fezf2 dosage shape cortical neuron development? 121,889 cells, with effect sizes calibrated against replicate noise. | bioRxiv |
- Every README lists the limitations and open issues.
- Code, software versions and data checksums are released with each project.
I founded CHIRAL Bangladesh and DeepBio to train students in reproducible bioinformatics. Course material lives at DeepBio Academy.

