AlphaFold Expands Beyond Proteins
Google DeepMind released a significant update to AlphaFold 3 this week, extending the model's structural prediction capabilities to RNA-protein interactions with accuracy that rivals experimental methods. The update is immediately available through the AlphaFold Server for non-commercial research use, with API access for pharmaceutical and biotech companies through Google Cloud.
Why RNA-Protein Interactions Matter
Proteins and RNA interact constantly in cellular machinery — ribosomes are RNA-protein complexes, many gene regulatory systems involve RNA binding proteins, and a growing class of therapeutics (including mRNA vaccines and RNA interference drugs) depend on understanding these interactions at the molecular level. Until now, experimental determination of RNA-protein complex structures required months of crystallography or cryo-EM work. AlphaFold 3's updated model predicts these structures in seconds.
Accuracy Benchmarks
On the RNA-Puzzles benchmark, AlphaFold 3's updated model achieves median TM-score of 0.87 for RNA-protein complexes, compared to 0.63 for the best prior computational methods. For free-standing RNA structures (not in complex with protein), performance is slightly lower at 0.81, reflecting the greater conformational flexibility of RNA compared to protein.
Research Applications
Early adopters include groups at Harvard Medical School using the tool to understand how mutation-disrupted RNA binding causes amyotrophic lateral sclerosis (ALS), and pharma companies developing RNA-targeting small molecule drugs. The commercial applications in RNA therapeutics alone represent a multi-billion dollar research acceleration opportunity.
Open Science Commitment
DeepMind continues to provide AlphaFold Server access free for academic research, a policy that has made it the most widely used structural biology tool globally.