← Back to index

Sep 2025 – Present

Undergraduate Research Assistant, Data Science (Protein Modeling)

The Pennsylvania State University · University Park, PA

RESEARCH · CURRENT

0.43 → 0.67
Spearman ρ
5
embedding strategies benchmarked
  • The lab is trying to predict how single point mutations change an enzyme's kinetic parameters (Km and kcat) for human acetylcholinesterase, using only sequence information, a hard regression problem when labeled data is scarce.
  • Pulled curated kinetics from SABIO-RK and generated ESM2 protein-language-model embeddings for wild-type and mutant sequences.
  • Benchmarked 5 different ways of turning those embeddings into a per-mutation feature, including raw difference vectors and kernel-based similarity metrics, raising held-out Spearman correlation from 0.43 to 0.67.
  • Implemented the full pipeline (featurization, training, evaluation) in Python with scikit-learn; the project is still active.
← Back to index