Research direction

AI for Chemistry and Materials Science

Machine learning systems for chemistry and materials science, with emphasis on molecular representations, catalysis, materials characterization, and domain-aware scientific models.

Chemistry Materials science Molecular representation Catalysis

Focus

This direction develops AI methods for chemistry and materials science where structure, tokens, measurements, and physical constraints all matter. The work connects representation learning, scientific language models, physics-informed fitting, and optimization for molecular and materials workflows.

Typical Questions

  • How should models combine molecular structure, text, and experimental measurements?
  • How can AI support chemistry when data is expensive, heterogeneous, or sparse?
  • How can domain knowledge and physical constraints improve scientific predictions?
  • Which representations best support materials discovery and chemical reasoning?
Selected papers

Chemistry and Materials Highlights

5 papers
Molecular graph and chemical tokens fusing into a joint chemistry embedding
2026 · Chemistry / Representation Learning

Thinking like a CHEMIST: Combined Heterogeneous Embedding Model Integrating Structure and Tokens

  • Combines molecular structure and token-based information in one heterogeneous embedding model.
  • Targets chemistry workflows where symbolic and structural signals are both important.
  • Frames chemical reasoning as representation learning across complementary modalities.
Paper PDF
Band-excitation microscopy signal fitted by a physics-constrained neural network
2023 · NeurIPS AI for Materials

Rapid Fitting of Band-Excitation Piezoresponse Force Microscopy Using Physics Constrained Unsupervised Neural Networks

  • Uses an unsupervised neural network constrained by a known empirical equation.
  • Targets real-time, robust fitting for high-dimensional scanning probe spectroscopy data.
  • Replaces brittle post-hoc analysis with physics-informed learning.
Paper PDF
Magnetics papers and a knowledge graph feeding a language model that extracts Curie temperature
2024 · Materials / Language Models

MagBERT: Magnetics Knowledge Aware Language Model Coupled with a Question Answering Pipeline for Curie Temperature Extraction Task

  • Adapts language-model tools to scientific information extraction in magnetics.
  • Combines domain knowledge with a question-answering pipeline for Curie temperature extraction.
  • Illustrates AI for science as structured extraction, not just generic prediction.
Paper PDF
Catalyst surface and methane chemistry experiments training a neural network for yield and selectivity
2022 · Catalysis / High-Throughput Chemistry

A deep neural network for oxidative coupling of methane trained on high throughput experimental data

  • Uses high-throughput experimental data for a catalytic chemistry task.
  • Applies neural modeling to oxidative coupling of methane.
  • Connects chemistry, energy, and data-driven materials discovery.
Paper PDF
Methylated DNA marks shifting transcription factor binding beyond canonical CpG contexts
2025 · PNAS / Genomics

DNA methylation shapes transcription factor binding beyond canonical CpG contexts

  • Studies how cytosine methylation changes transcription factor-DNA binding behavior.
  • Shows methylation effects extend beyond canonical CpG contexts and can reshape binding affinity.
  • Connects machine learning, synthetic DNA libraries, and regulatory genomics.
Paper PDF