Generative AI for molecular design, decoding 3D genome folding, and a CVPR invited talk

Autoregressive fragment-based diffusion for pocket-aware ligand design. NeurIPS GenBio Workshop (2023).

Mahdi developed AutoFragDiff, a diffusion model that generates 3D molecular fragments directly inside protein binding pockets. Rather than designing whole molecules at once, the model builds them piece by piece, improving local geometry while maintaining predicted binding affinity.

In silico discovery of repetitive elements as key sequence determinants of 3D genome folding. Cell Genomics (2023).

Laura and collaborators used deep learning to evaluate how DNA sequence changes affect chromatin organization across the genome, discovering that repetitive elements can rival or exceed CTCF sites in their effects on nearby chromatin interactions.

Michael gave an invited talk at the CVPR CVMI Workshop on decoding hidden signal from neurodegenerative drug discovery high-content screens.