AlphaFold2
AlphaFold2 is an advanced AI model developed by DeepMind for predicting the 3D structures of proteins, nucleic acids, and small molecules with high accuracy. It utilizes diffusion techniques and supports multi-chain predictions to enhance folding accuracy, providing insights into the interactions of all biological molecules, including ligands.
Inputs for Protein Structure Prediction
- Sequence: Provide the protein, RNA or DNA sequence in FASTA format.
- Templates: Include any available template structures (PDB files) to enhance accuracy.
- Features: Specify relevant features such as known mutations, post-translational modifications, and secondary structure information.
- Environment Settings: Define the computational environment, including hardware specifications and parallel processing options.
Outputs from AlphaFold 2
- PDB File: A file containing the predicted 3D coordinates of each atom in the input, providing a detailed representation of the molecular structure.
- pLDDT Scores: Specified in the b-factor column of the PDB file, these per-residue confidence scores range from 0 to 100, with higher scores indicating greater confidence in the prediction.
Analyzing Alphafold2 Predictions
Analyzing AlphaFold 2 predictions involves assessing predicted protein structures for accuracy and reliability. Key elements include evaluating pLDDT scores for structural confidence and interpreting PAE scores for residue dynamics.
pLDDT Scores
pLDDT scores provide a per-residue confidence metric indicating the reliability of the predicted folding. Scores from 0-50 indicate very low confidence, while scores from 90-100 signify high structural stability.
PAE Scores
PAE scores evaluate the confidence in the relative positions of residue pairs, with lower scores (0-5 Å) suggesting known positions and higher scores (20+ Å) indicating uncertainty in their spatial arrangement. This information is vital for understanding folding dynamics.