Synergic entanglements between Structural Biology and Machine Learning: Looking into the future

The rapid advances in machine learning (ML) and artificial intelligence (AI) are transforming biology and opening new directions for scientific inquiry. AI-based models are now routinely used in structure determination, yet they also raise scientific and societal challenges requiring timely attention. The combination of ML and structural biology is enabling the structural elucidation of proteins and their complexes, structure-based function annotation, prediction of mutation effects on protein function in disease, and integrative modelling of large cellular machines - with potential to drive effective treatments for conditions such as infection, cancer, and dementia.

From a machine-learning standpoint, structural biology provides a uniquely challenging arena that is driving methodological innovation in its own right. Biomolecular data can for instance demand architectures that respect physical symmetries and constraints - spurring advances in geometric and equivariant deep learning, SE(3)-invariant networks, and structure-conditioned diffusion models. At the same time, the scarcity and heterogeneity of experimental data can call for progress in data-efficient learning, transfer learning across protein families, and multi-modal integration of complementary experimental signals (e.g., crystallography, cryo-EM density maps, cross-linking mass spectrometry, and NMR observables). Equally pressing are the needs for reliable uncertainty quantification - critical when predictions inform drug-design decisions - and for interpretable models whose outputs can be interrogated in mechanistic terms.

This Research Topic explores what is currently achievable at the intersection of ML and structural biology, from modelling dynamic conformational states and advancing structural elucidation, to AI-assisted drug design, and how these developments may shape biology over the longer term. We welcome contributions across, but not limited to:

• Active-learning strategies for experiment prioritisation
• Generative models for conformational ensembles
• Self-supervised pre-training on structural databases
• Benchmarking frameworks that rigorously assess generalization to unseen folds and novel complexes
• Biophysical methods that quantify motion and heterogeneity
• Quantitative interaction biology
• Misfolding, aggregation, and assembly pathways
• In vivo / cell-like macromolecular environments

We aim to discuss the synergies, challenges, and frontiers of these fields, with particular attention to how ML may broaden access to structural biology globally. Long recognized as a powerful route to drug discovery and vaccine development, structural analysis has historically required expensive infrastructure and intensive manual effort. AI-based predictions, combined with remote synchrotron data collection, innovations in cryo-electron microscopy, and automated structure-determination pipelines, are now bringing these capabilities within reach of the wider international scientific community and we will address how this shift can benefit science across geographic and economic boundaries, with direct consequences for human health.