field notes · computational biology · est. nyc → cambridge
Yiyang (Steven) Yu俞翌阳
Incoming PhD student in Computational & Systems Biology at MIT. Steering reality toward a future where we can program biology as easily and accessibly as we program computers.
- 01github
- 02scholar
- 03x / twitter
- 04linkedin
- 05yiyu@mit.edu
↓ scroll · click the water to gather the fish
About
I studied Biomedical Engineering and Computer Science at Columbia. I work where deep learning meets biology, using one to accelerate discovery in the other.
Previously, I did research with Dr. Mohammed AlQuraishi at Columbia University Irving Medical Center and Dr. Peter Koo at Cold Spring Harbor Laboratory, building protein and DNA language models. On the side, I train neural networks to solve problems across all kinds of fields as a Kaggle Competitions Master, and build tools people want, like zotero-mcp.
In my free time, I love swimming, playing racket sports, cooking, and capturing moments through my lens.
fieldnote № 1 — why the fish?
My Chinese name is 俞翌阳, pronounced exactly like 鱼一羊 — "fish, one, sheep." The ocean you're looking at is my name, swimming. (The sheep regrettably could not be animated.)
Log
- 2026.04New preprint on D3: DNA Discrete Diffusion for regulatory sequence design, published on bioRxiv!
- 2026.04Awarded the NSF Graduate Research Fellowship!
- 2026.03Committed to the PhD program in Computational & Systems Biology at MIT!
- 2025.07Paper on DNA language models for regulatory genomics published in Genome Biology!
- 2025.05Paper on mechanistic interpretability of protein language models published at ICML!
- 2025.03Got named a 2025 Goldwater Scholar!
- 2025.01Won 3rd place in the Kaggle Santa 2024 competition!
Selected Work
-
α
research
D3: Designing DNA with Discrete Diffusion
A discrete diffusion model that designs regulatory DNA with tunable activity. I built the latent-space visualizer that shows how it shapes sequences and their motifs across diffusion steps.
-
β
tool
Zotero MCP
An MCP server that connects your Zotero library to Claude, ChatGPT, and other AI assistants, with semantic search over your papers. It has grown past 4,000 stars.
-
γ
tool
Nano Protein Viewer & MolView
Two lightweight molecular viewers built on Mol*. One lives in VS Code and the browser, the other is a Jupyter widget for notebooks.
-
δ
competition
Leash BELKA: Predicting New Medicines
A gold-medal Kaggle solution (13th of 1946) for predicting small molecule to protein binding. We blended over 40 deep-learning models, from GNNs and transformers to gradient-boosted trees.
Contact
For collaboration, research, or arguments about whether is all you need.