All publications

ProCyon is an 11B-parameter model that predicts and generates protein phenotypes across molecular and therapeutic scales, enabling transfer to poorly characterized proteins.

We survey graph machine learning for biomedical applications, outlining how relational models and emerging foundation models can deliver clinically meaningful predictions.

TimeX learns interpretable surrogate masks that mirror predictor behavior through a novel consistency loss, delivering faithful explanations for time-series models.

RAINCOAT introduces a domain adaptation framework tailored to time series, achieving state-of-the-art performance even when shifts appear in both features and labels.

GraphXAI benchmarks GNN explainers with new datasets, metrics, and evaluation flows that now inform tooling such as PyTorch Geometric.

We design a GNN architecture for polymer property prediction, achieving state-of-the-art performance and enabling high-throughput materials search.

We fine-tune protein language models on microbial genomics data and pair them with structure-aware explanations to reveal functional differences between surface and subsurface proteins.

Using motif-level genomic summaries, we predict poplar tree locations without full genome alignment, matching the performance of whole-genome approaches.

LASSO-driven feature selection reliably identifies informative genomic features for sepsis prediction and environmental profiling in microbial datasets.