Research
研究方向 · 理念 · 进行中的工作
Research Interests
研究方向
AI for Science
Leveraging machine learning to accelerate scientific discovery in chemistry and biology. I'm interested in building models that not only predict but also reveal underlying scientific principles.
Molecular Representation Learning
Learning effective and transferable molecular representations from graph, sequence, and 3D structural data. Exploring how different molecular views complement each other.
Mass Spectrometry
Computational methods for tandem mass spectrometry data, including spectrum prediction, metabolite identification, and de novo structure elucidation from MS/MS data.
Knowledge-guided Machine Learning
Incorporating chemical fragmentation rules, biological priors, and physical constraints into deep learning to make models more data-efficient, interpretable, and scientifically grounded.
Scientific Foundation Models
Developing and adapting large-scale pre-trained models for scientific data — from molecular graphs and sequences to mass spectra — enabling cross-modal understanding and knowledge transfer.
Research Philosophy
研究理念
I am interested in building machine learning systems that can reason with scientific knowledge instead of relying solely on statistical correlations. My long-term goal is to develop foundation models for scientific data that integrate domain knowledge from chemistry and biology, enabling more reliable molecular understanding and discovery.
I believe the most impactful AI for Science research sits at the intersection of domain expertise and machine learning innovation — understanding the science deeply enough to know what questions to ask, and building the right models to answer them.
Ongoing Work
进行中的研究
These are directions I am actively exploring. Some may lead to publications, others are early-stage explorations — all reflect where my thinking is heading.
Knowledge-guided Molecular Retrieval
Exploring how chemical fragmentation mechanisms and biological knowledge can be incorporated into retrieval models for metabolite identification from tandem mass spectra.
Chemical Fragmentation Modeling
Building models that understand how molecules fragment in mass spectrometers, integrating physical rules with learned representations for better spectrum prediction.
Scientific Reasoning for Mass Spectrometry
Developing methods that combine learned representations with explicit reasoning over chemical structures and fragmentation pathways for more interpretable metabolite identification.
Foundation Models for Molecular Science
Exploring cross-modal foundation models that jointly understand molecular graphs, sequences, spectra, and textual descriptions for comprehensive molecular understanding.