title: Boltz weight: 30

Boltz

General Use

Boltz mode can be run using the command below:

nextflow run nf-core/proteinfold \
    --input samplesheet.csv \
    --outdir <OUTDIR> \
    --mode boltz \
    --boltz_db <PATH> \
    --colabfold_db <PATH> \
    --use_gpu \
    -profile <docker/singularity/podman/shifter/charliecloud/conda/institute>

By default, --mode boltz will generate MSA files required for structure prediction using a local execution of the ColabFold search protocol. This protocol uses MMseqs2 to search a uniref30 expandable profile database and construct paired alignments using taxonomic labels. MSAs are enriched with additional unpaired sequences by searching an expandable profile databased of environmental sequences.

File Structure

The file structure of --boltz_db must be as follows:

<boltz_db>/
└── params
    ├── boltz1_conf.ckpt
    ├── boltz2_aff.ckpt
    ├── boltz2_conf.ckpt
    ├── ccd.pkl
    └── mols

If individual components are available at different locations in the filesystem, they can be set using the following flags:

# Boltz-1
--boltz_ccd_path <PATH/TO/ccd.pkl>
--boltz_model_path </PATH/TO/boltz1_conf.ckpt>
 
# Boltz-2
--boltz2_aff_path </PATH/TO/boltz2_aff.ckpt>
--boltz2_conf_path </PATH/TO/boltz2_conf.ckpt>
--boltz2_mols_path </PATH/TO/mols/>

Similarly, the --colabfold_db flag must be set to run the local execution of ColabFold search. The file structure of --colabfold_db must be:

<colabfold_db>/
├── colabfold_envdb
│   ├── colabfold_envdb_202108_db
│   ├── colabfold_envdb_202108_db_aln
│   ├── colabfold_envdb_202108_db_aln.dbtype
│   └── ...
└── colabfold_uniref30
    ├── uniref30_2302_db
    ├── uniref30_2302_db_aln
    ├── uniref30_2302_db_aln.dbtype
    └── ...

Without setting the --boltz_db and --colabfold_db flags, all of the required data files will be downloaded during the workflow execution.

As an alternative, Boltz MSAs can be generated without downloading the large reference sequence databases by calling the public MMSeqs API with the --use_msa_server argument. Users can also point to a private api endpoint using the --msa_server_url argument.

nextflow run nf-core/proteinfold \
    --input samplesheet.csv \
    --outdir <OUTDIR> \
    --mode boltz \
    --boltz_db <PATH> \
    --use_msa_server \
    --use_gpu \
    -profile <docker/singularity/podman/shifter/charliecloud/conda/institute>

General Molecules

Boltz can support general molecular structure prediction. The most direct way to indicate molecular type is to format FASTA files with the molecular type indicated in the sequence header:

>A|protein
QLEDSEVEAVAKGLEEM
>B|rna
AUGC
>C|smiles
N[C@@H](Cc1ccc(O)cc1)C(=O)O
>D|ccd
ATP
>E|dna
ATGC

If the molecule type is not specified in the header of the input fasta, proteinfold will try to guess the expected molecule type based on the character composition.

YAML format

Boltz allows specifying post-translational modifications and manual distance constraints to guide predictions. However, this input information is not supported in the FASTA format and must be specified in an input YAML file according to the boltz specification.

Boltz YAML files can be run with proteinfold in boltz mode by substituting the typical FASTA file in the input samplesheet.

id,fasta
T1024,T1024.yaml

Additional Arguments

See the Boltz documentation for a full description of additional arguments. The arguments supported by the proteinfold workflow are described briefly below:

You can override any of these parameters via the command line or a params file.