Chief Scientific Officer, Enrich Biosystems · Ph.D., Tsinghua University
I work where immunology, protein design, and measurement meet. My background is in physics and mathematics, and most of what I do is turning wet-lab experiments into models that make falsifiable predictions — then designing the next experiment to test them. Named inventor on the patents behind the TROVO single-cell microwell platform.
Two threads run through everything here:
- Measure single cells, not populations. Bulk cytotoxicity assays average away the thing you care about. Imaging effector–target pairs in individual microwells turns a killing curve into a distribution of per-cell rates, which is where the biology is.
- Make the modeling assumption explicit. Whether it's a two-component CAR⁺/CAR⁻ mixture, a predator–prey ODE, or a header-annotation rule for parsing PDB chains, the choice should be written down and defensible — not buried in a script.
| Repository | What it is |
|---|---|
trovo-analysis |
CAR-T and TIL killing kinetics from microwell co-culture imaging: EM mixture model for per-well k_eff, CARRGO-style predator–prey ODE fits, single-effector-cell inter-kill intervals, and a C++17 reimplementation of the fitter for batch runs |
fusiondesign-bglobin |
mRNA neoantigen vaccine design: epitope scanning and MHC binding, fusion construct assembly, HBB UTR/Kozak/poly-A build, IVT template output |
openlab |
Lab automation from YAML-defined protocols — 96/384-well plasmid prep, sandwich ELISA, DNA extraction, VHH expression and nickel purification — with hardware config and BOM |
nblib |
Structural nanobody library mined from the PDB: RCSB search, nanobody/antigen chain splitting, renumbering and superposition |
vscreen |
Antibody/nanobody repertoire NGS pipeline: FASTQ QC → IgBLAST → clonal clustering → consensus reconstruction, parallel at the annotation and consensus stages |
pymol-scripts |
Structure-prep utilities for design workflows: chain renumbering and merging (with and without chain-break gaps), interface hotspot shell selection |
Not public: immune-repertoire pipelines, binder design and docking campaigns, and protein engineering work that is either unpublished, covers pending IP, or runs on collaborator data. Happy to talk through any of it.
These repositories were consolidated from several years of project directories and published in August 2026 — the commit history starts there even where the work does not.
No experimental data is included. Raw plates, patient-derived samples, and collaborator datasets stay out of version control; notebook outputs are stripped before commit; and example sample names are generic. What's published is the method, not the findings of any particular study.