
Diffusion models for 3D molecular generation
Investigate equivariant diffusion architectures for generating 3D conformers with target binding properties.
For students
A selection of project ideas for master's theses, internships, and research collaborations. These are starting points — feel free to propose your own direction.

Investigate equivariant diffusion architectures for generating 3D conformers with target binding properties.

Build an end-to-end pipeline that extracts entities and relations from biomedical literature into a queryable graph.

Benchmark message-passing variants on heterogeneous biomedical graphs and propose improvements.

Design a conditional generative model that respects user-specified physicochemical constraints.

Explore contrastive and masked-modeling objectives to pretrain transferable molecular representations.

Develop visualization and attribution techniques to make generative model decisions interpretable to chemists.