AI model used to generate complete models of proteins in motion

2026-06-26 · Source: ΑΙhub · Field: Science & Research — Artificial Intelligence & Machine Learning, Life Sciences & Biology, Health & Medical Research · Depth: Expert, short

Summary

Scientists at EPFL's School of Life Sciences and Engineering have developed Latent Diffusion for Full Protein Generation (LD-FPG), an AI-based generative framework that produces complete, all-atom structural ensembles of proteins and their movements. Unlike existing systems such as Google DeepMind's AlphaFold, which generate static protein "snapshots", LD-FPG captures subtle atomic rearrangements in side chains, crucial for understanding protein interactions. Led by Patrick Barth and Pierre Vandergheynst, the framework uses a graph neural network (GNN) to compress protein structure data into a low-dimensional latent map, which an AI model then learns to generate new dynamic structures. This method enables the generation of full range of motion for complex drug targets like G-protein coupled receptors (GPCRs), including the dopamine D2 receptor in both active and inactive states. Published in NeurIPS 2025, LD-FPG aims to accelerate drug discovery by improving virtual screening processes.

Key takeaway

For research scientists focused on drug and antibody discovery, LD-FPG offers a new paradigm for understanding protein behavior. You should consider integrating dynamic protein models into your virtual screening processes to move beyond static snapshots. This approach can accelerate the identification of drug candidates by targeting a protein's full range of motion, not just its shape, potentially improving success rates in drug development.

Key insights

LD-FPG is the first AI framework to generate complete, all-atom dynamic protein models, capturing full motion.

Principles

• Protein dynamics are crucial for drug interaction.
• Low-dimensional maps enable all-atom dynamics.
• High-quality data is essential for AI success.

Method

LD-FPG uses a graph neural network to compress protein data into a latent map, which an AI learns to generate new dynamic structures, then converts back to high-resolution proteins.

In practice

• Model G-protein coupled receptors (GPCRs) motion.
• Enhance virtual screening for drug discovery.
• Analyze dopamine D2 receptor states.

Topics

• Protein Dynamics
• Generative AI
• Drug Discovery
• Graph Neural Networks
• Structural Biology
• GPCRs

Best for: AI Scientist, Research Scientist

Related on AIssential

• From possibility to precision in macromolecular ensemble prediction — Predicting dynamic protein conformational ensembles requires integrating diverse experimental data and advanced AI.
• OpenAI debuts GPT-Rosalind, a new limited access model for life sciences, and broader Codex plugin on Github — GPT-Rosalind is a specialized AI model and ecosystem accelerating life sciences research through domain-specific reasoning and workflow integration.
• Local or Global? The Big Question in AI Drug Discovery | Isomorphic Labs — Drug discovery ML models can be local (small data, specific problem) or global (large data, generalizable).
• A diffusion model conditioned on compound bioactivity profiles for generating high-content images — pDIFF uses bioactivity profiles to generate high-content images, improving virtual hit expansion for drug discovery.
• AWS and Hopkins Engineering announce groundbreaking database for AI/ML antibody design — A new diverse, experimentally validated dataset enables rigorous benchmarking of AI models for antibody developability.
• Bringing AI-driven protein-design tools to biologists everywhere — Democratizing AI-driven protein design tools empowers biologists to accelerate drug discovery and engineer novel proteins.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ΑΙhub.