Cheng Zeng
Computational Materials Scientist at University of Florida.
Postdoctoral Associate
Email:c.zeng@ufl.edu or cheng_zeng1@alumni.brown.edu
Hi! I am currently a postdoc working with Prof. Mingjie Liu at University of Florida. My research focuses on molecular-graph generative models and AI-driven computational study on materials for energy applications. My research interests span a multitude of topics including materials design, machine learning, electrochemistry, generative modeling, computational catalysis and renewable energy. I am interested in solving complex engineering problems with machine learning, physical simulations and numerical methods.
From 2022 to 2024, I worked at the Instutite for Experiential AI and Roux institute of Northeastern University on highly interdisciplinary topics across computational materials science, additive manufacturing, wind energy and human-data interaciton, mentored by Prof. Nathan Post and Jack Lesko. Prior to joining Roux, I received a PhD in Engineering at Brown University. En route to my PhD degree, I also received a secondary Master’s degree in Data Science via Brown’s Open Graduate Education Program. At Brown, I worked at the Catalyst Design Lab under the direction of Andrew Peterson. My PhD research focused on machine learning potentials and force–displacement phenomenological models for rational catalyst design. Before Brown, I got a Bachelor and Master in Materials Science and Engineering from Tsinghua University. At Tsinghua, I worked with Yunhan Ling on oxide films/coatings resistant to hydrogen permeation.
Recent news [archive]
| 2025/07 | Our paper MolGuidance, a follow-up work of PropMolFlow, has been accepted in ICML 2025 Generative AI for Biology (GenBio) Workshop. In this work, we integrate three advanced property‐guidance strategies—classifier‐free guidance, autoguidance, and model guidance—into an SE(3)‐equivariant flow‐matching framework for conditional molecule generation. |
|---|---|
| 2025/05 | Our PropMolFlow framework—now available online—uses flow matching parameterized by an SE(3)-equivariant graph neural network for conditional molecular generation. We demonstrate competitive performance against state-of-the-art diffusion models across a range of molecular properties, while maintaining high structural validity, stability, and sampling efficiency. To support reproducibility and further research, we also release two datasets: (1) DFT-computed properties for molecules generated with PropMolFlow, and (2) an updated QM9 SDF file (based on the DeepChem release) in which all bond-order and charge inconsistencies have been corrected. |
| 2025/01 | I officially joined University of Florida Department of Chemistry as a postdoc to work on molecular-graph generative modeling and AI-driven computational simulations. |
