I'm Roshan. I use computation to study and improve biology, and I care about health, technology, and animals.
- 🎓 I'm a PhD student in Biological Engineering at MIT (2026–).
- 🧬 I most recently built PWPA, a model that predicts how cells respond to gene knockdowns no one has measured yet.
- 🔭 Before that, I co-developed Brieflow and co-founded the computational team at the Cheeseman Lab (Whitehead Institute, MIT).
| Project | What it does |
|---|---|
| PWPA (2026) | Prior-Weighted Partner Average: predicts responses to an unmeasured knockdown from knockdowns measured in the same cell context, plus gene–gene priors drawn from many datasets. Beats PIE on PIE's own benchmark (AUPRC 0.325 vs. 0.304) with less compute and more interpretable output. Built with an LLM-driven lab loop of ~580 runs. |
| Brieflow (2025) | Extensible pipeline for optical pooled screen data. Cut screen configuration time by 80% and module integration time by 93%. Nature Communications · Featured by Anthropic · MIT Prize for Open Data |
| Pycytominer (2025) | Functions for processing high-dimensional cell experiment readouts. Nature Methods |
| IDR Stream (2023) | Pipeline that turns Image Data Resource cell images into single-cell features. |
- Brieflow: an integrated computational pipeline for high-throughput analysis of optical pooled screening data. Nature Communications (2026)
- Reproducible image-based profiling with Pycytominer. Nature Methods (2025)
- Toward generalizable phenotype prediction from single-cell morphology representations. BMC Methods (2024), co-first author