cv

Graduate Research Assistant

Carnegie Mellon University

Conducting research on improving trustworthiness of large language models in chemistry and developing large-scale quantum chemistry datasets.

• Led MolErr2Fix benchmark project for evaluating LLM trustworthiness in chemistry
• Created multi-million-entry quantum chemistry dataset
• Collaborated with Isayev and CLAW research groups

Ph.D.

Carnegie Mellon University

AI for Science

M.S.

Carnegie Mellon University, School of Computer Science

Automated Science

• Machine Learning
• Active Learning
• Bioinformatics
• Automated Experimentation

Visiting Scholar

National University of Singapore, School of Computing

Visiting Scholar

• Deep Learning
• Neural Networks
• Robotics

B.S.

Huazhong Agricultural University

Zhang Zhidong Class - Advanced Class

• Deep Learning & Neural Network
• Database Systems
• Data Structures & Algorithms

MolErr2Fix: Benchmarking LLM Trustworthiness in Chemistry via Modular Error Detection, Localization, Explanation, and Revision

EMNLP 2025 (Oral Presentation)

Developed comprehensive benchmark for evaluating large language models' trustworthiness in chemistry through modular error detection, localization, explanation, and revision tasks.

Multi-View Representation Learning for Identification of Novel Cancer Genes and Their Causative Biological Mechanisms

Briefings in Bioinformatics

Integrated multi-omics data (genomic, transcriptomic, epigenomic, PPI) to identify 74 high-confidence candidate cancer genes.

HimGNN: Hierarchical Molecular Graph Representation Learning Framework for Property Prediction

Briefings in Bioinformatics

Designed hierarchical GNN with atom- and motif-level features and Transformer-based local augmentation module, improving performance across 8 benchmark datasets.

CHEM-AL: Active Learning for Molecular Property Prediction

Built active learning framework with uncertainty, QBC, and diversity sampling. Implemented Naïve Bayes and MLP (with MC Dropout) on MoleculeNet datasets.

• Achieved 50% reduction in labeling requirements versus random sampling

Computational Prediction of Protein–Protein Interactions (PPI)

Designed deep learning pipelines for SHS27K benchmark (16,912 interactions). Developed ProBERT-BiGRU-Attention and Transformer-based models.

• Achieved 83% accuracy and AUC 0.91
• Built GNN capturing structural context for unseen proteins

Large-Scale Multi-Property Quantum Chemistry Dataset

Created a multi-million-entry dataset including energies, forces, dipoles, quadrupoles. Benchmarked SchNet, PaiNN, E2GNN, Equiformer for molecular property prediction.

• Preparing publication and open dataset release