Research

A few directions define most of my current work. Each is anchored by representative papers; the full list is on the publications page.

AI for Science

Embedding atomic-level physical and chemical mechanisms—spectral properties, atom interactions, evolutionary sequence context—directly into deep networks, so models predict molecular properties and structures through physically grounded reasoning rather than superficial correlations.

  • ExplainBind: Explainable Physicochemical Determinants of Protein-Ligand Binding via Non-Covalent Interactions. Meng, Z.†, Bai, Z.†, et al., Oldham, W. M.*, Meng, Z.*, Xu, H.*, Loscalzo, J. bioRxiv, 2026.
  • EvoRMD: Integrating Biological Context and Evolutionary RNA Language Models for Interpretable Prediction of RNA Modifications. Wang, B., Zhang, H., Cui, T., Wang, X., Song, J.*, Xu, H.* Genome Biology, 2026.
  • How Well Can Off-the-Shelf LLMs Elucidate Molecular Structures from Mass Spectra Using Chain-of-Thought Reasoning? Wang, Y., Lu, W., Liu, L., Xu, H.*, Ling, H.* J. Cheminform., 2026.
  • TransPeakNet for Solvent-Aware 2D NMR Prediction via Multi-Task Pre-Training and Unsupervised Learning. Li, Y.†, Xu, H.†, Kumar, A., Wang, D., Heiss, C., Azadi, P., Hong, P.* Communications Chemistry, 2025.

AI4Health

Developing foundational models and explainable frameworks for multi-omics, medical imaging, and electronic health records (EHR)—individually and in multimodal combination—to advance clinical prediction and diagnosis.

  • Deep Learning-Based MRI Model for Predicting P53-Mutated Hepatocellular Carcinoma. Jia, L., Yang, Q., Jiang, H., Huang, G., Wang, Z., Guo, X., Li, J., Xu, H.*, Lei, J.* BMC Medical Imaging, 2025.

AI Fundamentals

Parameter-efficient fine-tuning and multimodal alignment methods that preserve fine-grained gradient sensitivities, so adaptation stays reliable, interpretable, and cheap to train at scale.

  • Not All Directions Matter: Toward Structured and Task-Aware Low-Rank Adaptation. Xiao, X., et al., Xu, H. ACL, 2026.
  • Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs. Xiao, X., et al., Xu, H. ECCV, 2026.

Chemistry

Earlier work on cobalt-catalyzed asymmetric radical reactions and total synthesis—the physical-organic-chemistry foundation that now informs how I build mechanism-aware AI models.

  • Enantioselective Radical N-Heterobicyclization by New Mode of Asymmetric Induction via Kinetically Stable Chiral Radical Center. Xu, H., Wang, D.-S., Zhu, Z.-Y., Deb, A., Zhang, X. P.* Chem, 2024.

Emerging Direction

Quantum Computing for Scientific Discovery — exploring where quantum algorithms can accelerate simulation and search in molecular and biological discovery pipelines. Publications forthcoming.