Yuhui Hong

The paper develops an agentic AI system using LLMs and domain tools to hypothesize and validate structures of unknown drug metabolites from mass spectrometry data.

FIDDLE is a deep learning method that predicts chemical formulas from tandem mass spectra, achieving high accuracy and speed compared to existing methods.

This paper introduces a task-specific transfer learning approach to improve small molecule retention time prediction with limited data, achieving higher accuracy across diverse experimental conditions.

This study develops confounder-free predictive models for host phenotype prediction using microbiome data, showing improved accuracy and biological interpretability.

The paper introduces MicroKPNN-MT, a model that predicts human phenotypes and missing metadata from microbiome data, improving disease prediction accuracy and model generalizability.

3DMolMS predicts tandem mass spectra from 3D molecular conformations, achieving high cosine similarity with experimental spectra and enabling generalized predictions across instruments.