Skip to content

Alphafold

Runs Alphafold model inference on MSA search results and generates 5 pdb foldings per input sample ranked by best score.

Alphafold

Quick Start

  1. Prepare MSA search results in A3M format.
  2. Select model and configuration options.
  3. Run the node to generate ranked protein structure predictions.

Setup Guide

1. Prepare Input Data

  1. Obtain MSA search results in A3M format (e.g., using the MSA Search node).
  2. Ensure input IDs are unique and correspond to your samples.

2. Configure Alphafold Node

  1. Choose model preset and relaxation options as needed.
  2. Set additional parameters (e.g., skip models, seed).

3. Run Prediction

  1. Execute the node to obtain ranked PDB foldings for each input.

Basic Usage

Single Sequence Prediction

  • Input a single A3M record to predict its structure.
  • Use default model and settings for standard monomer prediction.

Batch Prediction

  • Input multiple A3M records for batch structure prediction.
  • Each input will yield 5 ranked PDB outputs.

Configuration

Required Inputs

Field Description Type Example
a3m MSA search results in a3m format. A3M {"sample1": "..."}
search_templates Whether to search for available templates before prediction. BOOLEAN False
model_preset Which Alphafold model to run. COMBO monomer
models_to_relax Whether to run relaxation step (not supported yet). COMBO NONE
enable_gpu_relax Whether to run relaxation step on GPU or CPU. BOOLEAN True
skip_models Which models to skip (comma-separated indices 1-5). STRING 3,5
seed Base seed for randomness. INT 42
mode Mode to run the node in (MOCK, PROD, TEST). COMBO PROD

Optional Inputs

None

Outputs

Field Description Type Example
folding.pdb Ranked foldings as a dict {pdb_id_rank_id: pdb_content}. PDB {"sample1_ranked_0": "..."}

Best Practices

Input Preparation

  • Ensure A3M input records are correctly formatted and IDs are unique.
  • Use the MSA Search node for compatible input generation.

Model Selection

  • Use the default monomer model for standard predictions.
  • Only skip models if you need to speed up inference and understand the implications.

Troubleshooting

Common Issues

  • All models skipped: Cannot skip all 5 models; remove at least one from skip_models.
  • Relaxation not supported: Relaxation step is not implemented; set models_to_relax to NONE.
  • Multimer not supported: Multimer model is not available; use monomer presets.
  • Input ID mismatch: Ensure input IDs in A3M and FASTA match exactly.

Need Help?