Biophysics×Bioinformatics

Biophysics has mastered generating atomic-resolution structural ground truth via cryo-EM, while Bioinformatics has learned to predict protein structures at genome scale from sequence alone—yet the two communities still co-publish rarely. The shared bridge territories of protein folding, proteostasis, and amyloid aggregation are now pulling them into contact, because AI predictions are hitting an accuracy ceiling on disordered regions, complexes, and disease-relevant conformational states that only physical experimental data can break. The fusion creates a compounding feedback loop: bioinformatics scales prediction, biophysics corrects and enriches training signal, dramatically accelerating mechanistic drug discovery.

ColabFold crossed 10,000 citations in under two years—a democratization signal that sequence-to-structure prediction is now mainstream bioinformatics infrastructure. But the field has simultaneously exposed its limits: disordered regions, amyloid polymorphs, and multi-state complexes resist pure sequence-based prediction. On the other side, cryo-EM automation is generating micrograph data at volumes that outpace human analysis pipelines, creating urgent demand for bioinformatics-scale processing and annotation. Thirty authors already straddle both fields as solo contributors; the bridge fields of neurodegeneration and proteostasis are publishing at accelerating rates in both literatures. The intellectual and talent infrastructure for fusion is assembled—formal co-publication is the lagging indicator, not the leading one.

Structural biology groups with embedded ML/compute arms at major institutes (think EMBL-EBI, UCSF, MRC LMB-adjacent environments) that can co-locate cryo-EM wet-lab pipelines with bioinformatics annotation infrastructure. Computational labs that have already adopted cryo-EM density maps as validation constraints for ML models hold first-mover advantage in defining the new methodological standard. Groups working on neurodegenerative amyloid polymorphs are the highest-probability collision point, as they are forced to reconcile experimental heterogeneity with computational models simultaneously.

An open-source cryo-EM-to-ensemble annotation pipeline: a 12–18 month project building tooling that ingests motion-corrected micrograph stacks, runs automated particle classification at bioinformatics scale, and outputs conformational ensemble data natively linked to UniProt accessions and Ensembl variant coordinates—enabling direct, population-scale mapping of disease-associated variants to experimentally observed structural perturbations. Validate on an amyloid polymorph dataset where AlphaFold predictions are known to diverge from cryo-EM ground truth.