Quantifying the Role of OpenFold Components in Protein Structure Prediction

2025/11/06 by Hayes, Tyler L., Krishnan, Giri P.
Biochemistry, Genetics and Molecular Biology · #Artificial Intelligence (cs.AI) #Bioinformatics and Genomic Networks #Biomolecules (q-bio.BM) #FOS: Biological sciences #FOS: Computer and information sciences #Machine Learning in Bioinformatics #Protein Structure and Dynamics

paper · doi:10.48550/arxiv.2511.14781

openalex publication_date 2025/11/06 · openalex created_date 2025/11/23 · openalex updated_date 2026/07/28

Abstract

Models such as AlphaFold2 and OpenFold have transformed protein structure prediction, yet their inner workings remain poorly understood. We present a methodology to systematically evaluate the contribution of individual OpenFold components to structure prediction accuracy. We identify several components that are critical for most proteins, while others vary in importance across proteins. We further show that the contribution of several components is correlated with protein length. These findings provide insight into how OpenFold achieves accurate predictions and highlight directions for interpreting protein prediction networks more broadly.

Citations

• Advances in Protein Representation Learning: Methods, Applications, and Future Directions
• Interpreting and Steering Protein Language Models through Sparse Autoencoders
• InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders
• What Matters in Transformers? Not All Attention is Needed
• MLP Can Be A Good Transformer Learner
• AlphaFold two years on: validation and impact
• The geometry of hidden representations of large transformer models
• ExplainableFold: Understanding AlphaFold Prediction with Explainable AI
• BERTology Meets Biology: Interpreting Attention in Protein Language\n Models
• Continuous Automated Model EvaluatiOn (CAMEO) complementing the critical assessment of structure prediction in CASP12

Related

• AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination
• Protein Structure Prediction until CASP15
• Highly accurate protein structure prediction with AlphaFold
• Proteins with alternative folds reveal blind spots in AlphaFold-based protein structure prediction
• A structural biology community assessment of AlphaFold2 applications