AlphaFold3 for Molecule Docking
AlphaFold3 (aka AF3) is an advanced AI model developed by DeepMind for predicting the 3D structures of proteins, nucleic acids, small molecules, and post-translational modification with high accuracy. It supports multi-chains and therefore is also a cofolding or docking method.
How to Use AlphaFold3
Inputs
- Protein Sequence(s)
String(s) representing the single-letter amino acid sequence, consisting of the common 20 amino acids.
- RNA Sequence(s)
- String(s) representing the single-letter nucleotide sequence.
- DNA Sequence(s)
- String(s) representing the single-letter nucleotide sequence.
- SMILE(s)
- String representing a small molecule.
Outputs
- PDB File
- Specifies the predicted structure of the inputs.
Example
Due to the licensing limitations of AlphaFold3, Copilot serves an alternative, called DiffDock.
Dock asparagine into A0A1L9RXX7
Analyzing Alphafold3 Predictions
pLDDT Scores: The predicted local distance difference test (pLDDT) is a per-residue confidence score. Scores range from 0-50 (very low, indicating disorder/flexibility) to 90-100 (very high, suggesting highly structured/stable conformations). Understanding these scores is crucial for assessing the reliability of the predicted structure.
PAE Scores: Predicted aligned error (PAE) measures confidence between residue pairs. Scores of 0-5Å indicate high confidence (known relative positions), while scores of 20Å+ indicate low confidence (unknown relative positions). Analyzing PAE scores helps evaluate the structural alignment and potential flexibility of protein regions.