Hi! I am Kexin. I am a PhD student at Stanford Computer Science, advised by Prof. Jure Leskovec. My research is supported by Stanford Bio-X fellowship.

I work on enabling machine learning to produce novel, deployable, and interpretable biomedical discoveries. Questions that I am excited about:

  • How to encode the massive, multi-modal, and multi-scale biological experimental data as inductive bias and use them to generate novel hypotheses and discoveries?

  • How to make these discoveries reliable, trustworthy, and overall aligned with what scientists truly value? 

  • How to ground model predictions to first-principle mechanisms and actionable insights?

I answer these questions on diverse and difficult biological problems such as human genetics, genetic perturbations, therapeutics discovery, etc. 

Previously, I worked with Prof. Marinka Zitnik, Dr. Cao Xiao, and Prof. Jimeng Sun. I have spent time researching at Genentech, Pfizer, IQVIA, Dana-Farber, Flatiron Health, and Rockefeller University. I did my undergrad at NYU in math & CS & studio art, and master at Harvard in health data science.

News:

2023.12 Spotlight at NeurIPS 2023 and Oral at MLCB 2023!

2023.06 Excited to work with Prof. Aviv Regev and Dr. Romain Lopez at Genentech this summer!

2022.07 Co-organize the inaugural LoG conference!

2022.07 Best paper honorable mentions award at IEEE VIS!

2022.07 Co-organize AI for Science Workshop at ICML!

2021.07 Co-organize AI for Science Workshop at NeurIPS!

Research:

[ Preprint ] Zero-shot Prediction of Therapeutic Use with Geometric Deep Learning and Clinician-Centered Design

[ Preprint ] Relational Deep Learning: Graph Representation Learning on Relational Databases

[ RECOMB ] Sequential Optimal Experimental Design of Perturbation Screens Guided by Multimodal Priors

[ NeurIPS ] Uncertainty Quantification over Graph with Conformalized Graph Neural Networks

[ NeurIPS ] Enabling Tabular Deep Learning When d≫n with an Auxiliary Knowledge Graph

[ Nature Biotechnology ] GEARS: Predicting Transcriptional Outcomes of Novel Multi-gene Perturbations

[ Nature ] Scientific Discovery in the Age of Artificial Intelligence

[ Nature Chemical Biology ] Artificial Intelligence Foundation for Therapeutic Science

[ Nature Biomedical Engineering ] Graph Representation Learning in Biomedicine and Healthcare

[ NeurIPS ] Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development

[ IEEE VIS ] Towards Usable Explanations: Extending the Nested Model of Visualization Design for User-Centric XAI

[ NeurIPS ] Graph Meta Learning via Local Subgraphs

Visit here for more.