1 Overview | 2 Using | 2.1 Version alphafold3/3.0.0

Overview

GitHub: alphafold3: AlphaFold 3 inference pipeline.

Using

Version alphafold3/3.0.0

The example below uses the following folder structure:

input_data/
  input.json
  input.fasta
output_data/

Create the alphafold3 input.json file

One of the main differences between alphafold and alphafold3 is that the latter defines an input.json file.

This can be created from an input.fasta file using the af3cli tool:

[]# module load af3cli/0.4.2
[]$ cd input_data
[]$ cat input.fasta
>NEKL-3 / F19H6.1 ATP-binding (29-52) active site (146-158)
MDKISNIYNFDDPPPDKLSLELFIIEKKIGKGQFSEVFRAQCTWVDLHVALKKIQVFEMVDQKARQDCLKEIDLLKQLNHVNVIRYYASFIDNNQLNIVLELAEAGDMSRMIKHFKKGGRLIPEKTIWKYFVQLARALAHMHSKRIMHRDIKPANVFITGNGIVKLGDLGLGRFFSSKTTAAHSLVGTPYYMSPERIQESGYNFKSDLWSTGCLLYEMAALQSPFYGDKMNLYSLCKKIENCEYPPLPADIYSTQLRDLVSRCILPEASKRPETSEVLQVAEHMNNYFSPSGDQSTTPSTQF
>MLT-3 SAM Dimerization
MSFRVRFSQERPGVTYVPSSRSGNTDLFNDIFSKPMPIIRNEDHVLMDACTAASIGDENKLKELMKINPNTMIMKNHDGWTPLLYAAYLGHNSVAAFLLDNGAKVDDSTESRWQTPLMMASACGNISVVRLLLGRGANPRLCDKEKRQAIHYASSCCQNVVVDTLLAAGCDPNAADSEGTTPVHEAAMAGHEITFIALLEKGGNLELKNAKGENAAALACDNNKILQLISEHQAEIAKQSSKTGNRGPRRLSELLEEMDLSRYIDQFKHENVDLEVFFELKEQDFVDMNIAYGPKKRMLDVIERYKKTGVIRSDAFDAPQGTAMSTKGSSTRESSEKTTATLKSIKELNQETKKSVMTALESLGSGQDKVRAQLISILNNVETISLKVSSHV

[]$ af3cli fasta input.fasta
[]$ cat input.json
{
    "name": "job",
    "version": 1,
    "dialect": "alphafold3",
    "modelSeeds": [
        1
    ],
    "sequences": [
        {
            "protein": {
                "id": [
                    "A"
                ],
                "sequence": "MDKISNIYNFDDPPPDKLSLELFIIEKKIGKGQFSEVFRAQCTWVDLHVALKKIQVFEMVDQKARQDCLKEIDLLKQLNHVNVIRYYASFIDNNQLNIVLELAEAGDMSRMIKHFKKGGRLIPEKTIWKYFVQLARALAHMHSKRIMHRDIKPANVFITGNGIVKLGDLGLGRFFSSKTTAAHSLVGTPYYMSPERIQESGYNFKSDLWSTGCLLYEMAALQSPFYGDKMNLYSLCKKIENCEYPPLPADIYSTQLRDLVSRCILPEASKRPETSEVLQVAEHMNNYFSPSGDQSTTPSTQF"
            }
        },
        {
            "protein": {
                "id": [
                    "B"
                ],
                "sequence": "MSFRVRFSQERPGVTYVPSSRSGNTDLFNDIFSKPMPIIRNEDHVLMDACTAASIGDENKLKELMKINPNTMIMKNHDGWTPLLYAAYLGHNSVAAFLLDNGAKVDDSTESRWQTPLMMASACGNISVVRLLLGRGANPRLCDKEKRQAIHYASSCCQNVVVDTLLAAGCDPNAADSEGTTPVHEAAMAGHEITFIALLEKGGNLELKNAKGENAAALACDNNKILQLISEHQAEIAKQSSKTGNRGPRRLSELLEEMDLSRYIDQFKHENVDLEVFFELKEQDFVDMNIAYGPKKRMLDVIERYKKTGVIRSDAFDAPQGTAMSTKGSSTRESSEKTTATLKSIKELNQETKKSVMTALESLGSGQDKVRAQLISILNNVETISLKVSSHV"
            }
        }
    ]
}

Running Alphafold3

The bind commands below map your folders to those within the container image being run.

[]$ module load alphafold3/3.0.0

# The following environment variables are set:
AF3_PARAMS=/cluster/medbow/udata/alphafold3/PARAMETERS
AF3_DATA=/cluster/medbow/udata/alphafold3
AF3_SIF=/apps/u/opt/containers/alphafold3/3.0.0/alphafold3/alphafold3.sif
AF3_PY=/apps/u/opt/containers/alphafold3/3.0.0/alphafold3/run_alphafold.py
AF3_BASE=/apps/u/opt/containers/alphafold3/3.0.0/alphafold3

[]$ apptainer exec \
  --nv \
  --bind ./input_data/:/root/af_input \
  --bind ./output_data:/root/af_output \
  --bind $AF3_PARAMS:/root/models \
  --bind $AF3_DATA:/root/public_databases \
  $AF3_SIF \
  python $AF3_PY \
  --json_path=/root/af_input/input.json \
  --output_dir=/root/af_output \
  --model_dir=/root/models \
  --db_dir=/root/public_databases \
  --db_dir=/root/public_databases_fallback

Getting alphafold3 parameter details:

There are a lot more parameters available than those in the example above.

[]$ module load alphafold3/3.0.0
[]$ apptainer exec $AF3_SIF python $AF3_PY --help
AlphaFold 3 structure prediction script.
...