What is AlphaFold
Proteins are essential to life, and understanding their structure can facilitate a mechanistic understanding of their function. Only a fraction of unique protein structures are known out of billion of known protein sequences and accurate computational approaches are needed to address this gap and to enable large-scale structural bioinformatics.
AlphaFold is an AI system developed by DeepMind that predicts a protein’s 3D structure from its amino acid sequence. It regularly achieves accuracy competitive with experiment. To get started, user can read the AlphaFold example and adapt the SLURM script provided under /share1/alphafold/sample/ in HPC2021 system.
N.B. As AlphaFold was written to use only one GPU per job, request to SLURM for two or more GPUs in a single job would not result in shorter running time. Instead, it may result in longer queuing time as the submitted job will not run until multiple GPUs become available.
Module
To setup required environment variables for AlphaFold, please use following command
System | Version | Command |
---|---|---|
HPC2021 | 2.1.0 | module load anaconda/py3.8 cuda/11.2 alphafold/2.1.0 |
2.1.2 | module load cuda/11.2 alphafold/2.1.2 | |
2.3.1 | module load cuda/12.3 alphafold/2.3.1 |
Below Genetic (sequence) Databases are available at HPC2021 system under $AF_DB (/lustre1/software/alphafold/db):
- BFD,
- MGnify,
- PDB70,
- PDB (structures in the mmCIF format),
- PDB seqres – only for AlphaFold-Multimer,
- Uniclust30,
- UniProt – only for AlphaFold-Multimer,
- UniRef90
Citation
If you use this resource please cite the AlphaFold methods paper:
Jumper, J et al. Highly accurate protein structure prediction with AlphaFold. Nature (2021).