Cookbook

Chai-1

Chai-1 is a high accuracy, ligand-aware protein folding tool. It takes as input a FASTA file, and outputs a directory of folded PDB structures. Chai-1 scores are output as .npz files.

Simple

Fold a protein from a sequence

ribbon.Chai1(
        fasta_file = 'my_sequence.fasta',   # A single input FASTA. If there are multiple sequences, they will be folded in the same structure.
        output_dir = './out'                # Where the outputs will be stored
    ).run()

Advanced

Fold a protein from a sequence. Include two copies of ligand

ribbon.Chai1(
        fasta_file = 'my_sequence.fasta',   # A single input FASTA. If there are multiple sequences, they will be folded in the same structure.
        output_dir = './out',               # Where the outputs will be stored
        smiles_string = 'C1=CC=CC=C1',      # SMILES string of our ligand
        num_ligands = 2,                    # How many copies of our ligand?
        device = 'gpu'                      # Run on GPU (necessary for Chai-1)
    ).run()

LigandMPNN

LigandMPNN is a ligand-aware tool for designing sequences for a backbone structure. It takes as input a list of PDB files, and outputs a sequence (or sequences) that are predicted to fold into that structure. While LigandMPNN is gpu-accelerated, it seems to run fast on the CPU as well.

Simple

Design a sequence for a backbone:

ribbon.LigandMPNN(
        structure_list = ['my_structure.pdb'],  # List of PDB files
        output_dir = './out'                    # Output directory
        num_designs = 5                         # How many sequences should we generate?
    ).run()

The output folder will have the following structure:

output_dir/
├─ backbones/    # Backbone structures with labeled AAs (but no sidechains)
├─ packed/       # (Optional) Backbone structures with packed sidechains
├─ seqs/         # A single FASTA containing the reference sequence and all designed sequences
└─ seqs_split/   # Individual FASTA files for each designed sequence

Advanced

Design a homodimeric sequence, keeping crucial residues fixed. This example uses the extra_args parameter to add extra parameters into your run command. Note that this can inject arbitrary code into your container - use with caution!

ligandmpnn_task = ribbon.LigandMPNN(
        structure_list = ['my_structure.pdb'],  # List of PDB files
        output_dir = './out'                    # Output directory
        num_designs = 5                         # How many sequences should we generate?
        extra_args= '--fixed_residues \"' + RESIDUES_TO_KEEP + '\" --homo_oligomer 1'   # Make sure to keep my catalytric residues, and make two chains identical.
    )

RaptorX-Single

RaptorX-Single is a fast protein folding tool. It can fold small structures in as little as 5 seconds, after an initial loading period. It takes as input a FASTA file (or directory containing FASTAs), and outputs a directory of folded PDB structures.

Simple

Fold a directory of FASTA files

    ribbon.RaptorXSingle(
            fasta_file_or_dir = './my_FASTA_directory/',
            output_dir = './out'
    ).run()

Advanced

Fold a directory of FASTA files using a non-default model checkpoint (param). Run on the CPU (slower; GPU is default).

    ribbon.RaptorXSingle(
            fasta_file_or_dir = './my_FASTA_directory/',
            output_dir = './out',
            param = 'RaptorX-Single-ESM1b-ESM1v-ProtTrans-Ab.pt',
            device='cpu'
    ).run()

The available model parameters are:

'RaptorX-Single-ESM1b.pt',
'RaptorX-Single-ESM1v.pt',
'RaptorX-Single-ProtTrans.pt',
'RaptorX-Single-ESM1b-ESM1v-ProtTrans.pt',
'RaptorX-Single-ESM1b-Ab.pt',
'RaptorX-Single-ESM1v-Ab.pt',
'RaptorX-Single-ProtTrans-Ab.pt',
'RaptorX-Single-ESM1b-ESM1v-ProtTrans-Ab.pt'