AlphaESM hFolds

AlphaESM hFolds compares predicted protein structures from AlphaFold2 and ESMFold for 42,942 proteins from the Reference Human Proteome to enable model quality assessment against experimental PDB structures.

Key Features:

• Comparative Analysis: Provides paired AlphaFold2 and ESMFold models for each protein entry to enable direct structural comparison.
• Dataset Coverage: Includes 42,942 proteins from the Reference Human Proteome.
• Integration with Experimental Data: For 2,900 proteins, predicted models are aligned with experimentally determined structures from the Protein Data Bank (PDB) when structure-to-sequence coverage is at least 70%.
• Quality Assessment Metrics: Uses TM-score as the quantitative metric for model quality, with TM-scores above 0.6 indicating good agreement with PDB structures.
• Model Superiority Insights: Analysis indicates AlphaFold2 models generally achieve slightly higher TM-scores than ESMFold when superimposed on PDB structures, and approximately 55% of entries have models with TM-scores below 0.6.

Scientific Applications:

• Model Validation: Validate computational predictions against experimental PDB structures to assess accuracy.
• Methodological Comparison: Evaluate and compare the performance of AlphaFold2 and ESMFold across the human proteome.
• Protein Structure Prediction Research: Identify regions and proteins where predictions diverge from experimental structures to inform methodological improvements.

Methodology:

Curate a dataset of human proteins, integrate predicted models from AlphaFold2 and ESMFold, align those models with available PDB structures, and compute TM-scores for model quality assessment.