Designing Drugs
with Generative AI
Senior Research Associate at Takeda Pharmaceuticals — building GPU-accelerated AI pipelines for protein structure prediction, binding affinity modeling, and pocket-guided ligand design.
Research Focus
I'm a Senior Research Associate in Computational Drug Discovery at Takeda Pharmaceuticals, developing and deploying end-to-end AI pipelines for structure prediction, binding affinity modeling, and pocket-guided ligand design.
My work centers on integrating generative deep learning — diffusion models, equivariant graph neural networks, and transformer architectures — with structural biology and cheminformatics to address real therapeutic challenges. I partner with computational chemists and structural biologists to translate model outputs into deployable workflows and actionable design hypotheses.
Previously, I completed my MS in Data Science at Vanderbilt University, where I developed SuperMetal (NeurIPS MLSB 2024) and SuperWater, published in Communications Chemistry (Nature Portfolio). I hold a BS in Computer Science, Mathematics, and Data Science from the University of Wisconsin–Madison.
Research Career
- Build end-to-end GPU-accelerated pipelines for structure prediction, feature engineering, model training, and reproducible evaluation with automated experiment tracking.
- Fine-tune generative models for affinity prediction and pocket-guided ligand design; implement pose-quality filtering to improve downstream success rates.
- Develop a peptide design workflow integrating physics-based simulation with AI to incorporate noncanonical amino acid mutations.
- Automate high-throughput ligand and residue-mutation screening with ensemble model consensus for candidate prioritization.
- Use Schrödinger Maestro for pose inspection and LiveDesign to track structure–property relationships, streamlining handoffs with chemistry teams.
- Developed SuperMetal, demonstrating higher accuracy than AlphaFold 3 for metal ion position prediction — accepted to NeurIPS MLSB 2024; published in Journal of Cheminformatics (2025).
- Built SuperWater, a generative and geometric deep learning framework integrating score-based diffusion models with equivariant GNNs for protein water binding site prediction; published in Communications Chemistry, Nature Portfolio (Dec 2025).
- Designed a CNN-based method to detect antigen-binding sites by representing antigen surface information as 2D images (AUC > 0.93); published in BioSystems (2024).
- Implemented an active learning framework combining molecular dynamics simulations with ML to investigate peptide fibril formation; manuscript in preparation.
- Built an AI-driven web platform using Streamlit to transform academic papers into plain language summaries, with user authentication and content customization via Python and Deta.
- Head TA — DS 3262 Applied Machine Learning, Vanderbilt University (Fall 2024 – Spring 2025)
- TA — DS5220 Principles of Programming and Simulation, Vanderbilt University (2024–2025)
- TA — DS5640 Machine Learning, Vanderbilt University (2024–2025)
- Course Assistant — Math 535 Mathematical Methods in Data Science, UW–Madison (Spring 2022)
Academic Background
Research Output
Awards & Presentations
Interests
When I'm not building models or debugging pipelines, I find balance in creative pursuits. I've practiced Chinese calligraphy for over 14 years — the discipline of brushstroke mirrors the precision I bring to my research. Music is another passion; I self-teach guitar, discovering melodies one chord at a time. I'm always curious to explore something new that sparks creativity and keeps my thinking fresh.