Research

Four active directions, from method development to prospective experimental validation.

01

AI for antibody design

We design therapeutic antibodies with structure-aware machine learning and validate them prospectively in the lab.

AIntibody Challenge (top five of 166 teams) - AbBiBench affinity maturation benchmark - SimBinder-IF affinity-optimized inverse folding
Details →
AI for antibody design
Antibody design pipeline
02

LLM for scientific discovery

Large language models drive scientific discovery in data-scarce settings, from synthetic tabular data generation to drug synergy prediction.

MALLM-GAN: multi-agent LLM as generative adversarial network - CancerGPT: few-shot drug pair synergy prediction (top 10 most cited npj Digital Medicine paper, 2024)
Details →
LLM for scientific discovery
MALLM-GAN and CancerGPT
03

Biomedical graph modeling

Graph neural networks represent complex biomedical graphs and derive patterns for better understanding of human biology.

Knowledge graphs for drug repurposing - biological networks: cell-gene, protein-protein, drug-protein, brain networks
Details →
Biomedical graph modeling
Biomedical knowledge graphs
04

Counterfactual inference

We investigate counterfactual inference to identify causality among complicated variables observed in real-world patient data.

Heterogeneous treatment effect estimation benchmark - interpretable patient subgrouping from clinical trials
Details →
Counterfactual inference
Heterogeneous treatment effects

Funding

More than $8M in extramural funding as principal investigator over the past five years.

NIH R01AG084637
PI · 07/2024 - 06/2029
Data-driven Subtypes of Alzheimer's disease progression for targeted treatment

Identify subgroups of patients showing similar progression patterns across multimodal data, including cognitive function, brain atrophy, and biomarkers, by harmonizing multiple completed clinical trial data.

NIH R01AG082721
PI · 09/2023 - 05/2029
Harmonizing multiple clinical trials for Alzheimer's disease to investigate differential responses to treatment via federated counterfactual learning

Develop machine learning models to identify patient subgroups who respond differently to treatments, resulting in smaller, less expensive, and more targeted Alzheimer's disease clinical trials.

NIH U24AG098157
Co-I · 09/2025 - 08/2030
ReCARDO: Using Real-World Data to Derive Common Data Elements for Alzheimer's Disease and AD-Related Dementias

Establish a ten-institution research network and an AD/ADRD-focused collaborative data ecosystem incorporating common data elements and real-world data.

NIH R01HD120995-01
Co-I · 07/2026 - 06/2030
Defining, Validating, & Endotyping Phenotypes in BPD

Define and validate clinically and biologically meaningful subgroups of bronchopulmonary dysplasia in preterm infants using large-scale neonatal datasets and transcriptomic data.

NIH R01NR22258
Site PI · 04/2026 - 03/2030
An Artificial Intelligence Approach to Understanding Trade-offs in Emergency Department Decision-Making

Develop an evaluation framework for emergency department decision-making and use it to assess AI models designed to support rapid and accurate ED decisions.

Robert Wood Johnson Foundation
PI (completed) · 09/2019 - 09/2021
Computational Phenotyping to Better Understand Obesity

Develop a dynamic pattern mining model to identify trends and patterns of weight gain (temporal phenotyping) most likely to be diabetic.

NIH R01AG066749-03S1
MPI (completed) · 09/2022 - 08/2023
Finding combinatorial drug repositioning therapy for Alzheimer's disease and related dementias

Develop novel machine learning solutions to mitigate algorithmic unfairness by addressing data biases.