PROFOUND

Motion-Aware Foundation Models for Protein Design

PROFOUND is a pilot project of the Helmholtz Foundation Model Initiative (HFMI), a strategic initiative of the Helmholtz Association advancing foundation models for scientific research.

PROFOUND combines molecular dynamics, artificial intelligence, and high-performance computing to move protein design beyond static structures. By learning from molecular motion and conformational ensembles, we aim to develop foundation models and computational methods for understanding, predicting, and designing proteins in their dynamic, functional states.

PROFOUND is also recognised as a John von Neumann Institute for Computing (NIC) Excellence Project, supporting large-scale molecular simulations, AI model development, and systematic benchmarking on leading European supercomputing infrastructure, including JUPITER.

This Hugging Face organisation hosts open datasets, models, benchmarks, and evaluation tools developed within PROFOUND.

BILBY — Benchmarking Interactions, Landscapes, Backbone fidelity, and dYnamics

• BILBY-PRO — Protein motif scaffolding benchmark containing an explicit functional motif test set, derived from PROSITE.
• BILBY-MD — Molecular-dynamics-based evaluation framework to benchmark protein design engines and the conformational stability of their benchmarks.

Further datasets, foundation models, and tools will be released as they become available.

Project links

• PROFOUND — Helmholtz Foundation Model Initiative
• PROFOUND — NIC Excellence Project 2026/1
• Helmholtz Association