This repository is based on localcolabfold, and allows you to run ColabFold/AlphaFold2_advanced.ipynb locally.
It has some features that ColabFold and AlphaFold does not have.
It can output the all structures during the recycling process. For example, if 3 is specified for the recycle number, a total of three structures (1,2,3 recycles) will be output.
It is also possible to specify all models (model_[1-5] and model_[1-5]_ptm) or multiple ensembles (1 and 8).
Finally, aggregate the reliability scores (pLDDT and pTM) for all predicted structures and write them to a csv.
The following command will download the model parameters and create the miniconda environment for running AlphaFold.
bash install.shThe miniconda python environment will be created in colabfold-conda/bin/python3.7.
This will not overwrite the existing conda environment.
$ colabfold-conda/bin/python3.7 --help
usage: runner_af2advanced.py [-h] -i INPUT [-o OUTPUT_DIR] [-ho HOMOOLIGOMER]
[-m {mmseqs2,single_sequence,precomputed}]
[--precomputed PRECOMPUTED]
[-p {unpaired,unpaired+paired,paired}]
[-pc PAIR_COV] [-pq PAIR_QID]
[-b {pLDDT,pTMscore}] [--noranking] [-t]
[-mm MAX_MSA]
[--num_CASP14_models NUM_CASP14_MODELS]
[--num_pTM_models NUM_PTM_MODELS]
[--model {CASP14,pTM,both}]
[-e [{1,8} [{1,8} ...]]] [-r MAX_RECYCLES]
[--output_all_cycle] [--tol TOL] [--is_training]
[--num_samples NUM_SAMPLES]
[--num_relax {None,Top1,Top5,All}]
[--save_images] [--show_images]
Runner script that can take command-line arguments
optional arguments:
-h, --help show this help message and exit
-i INPUT, --input INPUT
Path to a FASTA file. Required.
-o OUTPUT_DIR, --output_dir OUTPUT_DIR
Path to a directory that will store the results. The
default name is 'prediction_<hash>'.
-ho HOMOOLIGOMER, --homooligomer HOMOOLIGOMER
homooligomer: Define number of copies in a homo-
oligomeric assembly. For example, sequence:ABC:DEF,
homooligomer: 2:1, the first protein ABC will be
modeled as a omodimer (2 copies) and second DEF a
monomer (1 copy). Default is 1.
-m {mmseqs2,single_sequence,precomputed}, --msa_method {mmseqs2,single_sequence,precomputed}
Options to generate MSA.mmseqs2 - FAST method from
ColabFold (default) single_sequence - use single
sequence input.precomputed - specify 'msa.pickle' file
generated previously if you have.Default is 'mmseqs2'.
--precomputed PRECOMPUTED
Specify the file path of a precomputed pickled msa
from previous run.
-p {unpaired,unpaired+paired,paired}, --pair_mode {unpaired,unpaired+paired,paired}
Experimental option for protein complexes. Pairing
currently only supported for proteins in same operon
(prokaryotic genomes). unpaired - generate seperate
MSA for each protein. (default) unpaired+paired -
attempt to pair sequences from the same operon within
the genome. paired - only use sequences that were
sucessfully paired. Default is 'unpaired'.
-pc PAIR_COV, --pair_cov PAIR_COV
Options to prefilter each MSA before pairing. It might
help if there are any paralogs in the complex.
prefilter each MSA to minimum coverage with query (%)
before pairing. Default is 50.
-pq PAIR_QID, --pair_qid PAIR_QID
Options to prefilter each MSA before pairing. It might
help if there are any paralogs in the complex.
prefilter each MSA to minimum sequence identity with
query (%) before pairing. Default is 20.
-b {pLDDT,pTMscore}, --rank_by {pLDDT,pTMscore}
specify metric to use for ranking models (For protein-
protein complexes, we recommend pTMscore). Default is
'pLDDT'.
--noranking Ranking output structures. If True, the output file
name contains a ranking.
-t, --use_turbo introduces a few modifications (compile once, swap
params, adjust max_msa) to speedup and reduce memory
requirements. Disable for default behavior.
-mm MAX_MSA, --max_msa MAX_MSA
max_msa defines: max_msa_clusters:max_extra_msa number
of sequences to use. This option ignored if use_turbo
is disabled. Default is '512:1024'.
--num_CASP14_models NUM_CASP14_MODELS
specify how many CASP14 model (normal model) params to
try. (Default is 5)
--num_pTM_models NUM_PTM_MODELS
specify how many pTM model params to try. (Default is
5)
--model {CASP14,pTM,both}
Model to use. CASP14 is a normal model. Both uses both
models. Number of models to be used is read from
num_CASP14_models and num_pTM_models respectively.
(Default is CASP14)
-e [{1,8} [{1,8} ...]], --num_ensembles [{1,8} [{1,8} ...]]
the trunk of the network is run multiple times with
different random choices for the MSA cluster centers.
(1=default, 8=casp14 setting)
-r MAX_RECYCLES, --max_recycles MAX_RECYCLES
controls the maximum number of times the structure is
fed back into the neural network for refinement.
(default is 3)
--output_all_cycle Output the structures of all cycle. If this option is
set and max_recycles is 3, the structures of all 1,2,3
cycles will be output.
--tol TOL tolerance for deciding when to stop (CA-RMS between
recycles)
--is_training enables the stochastic part of the model (dropout),
when coupled with num_samples can be used to 'sample'
a diverse set of structures. False (NOT specifying
this option) is recommended at first.
--num_samples NUM_SAMPLES
number of random_seeds to try. Default is 1.
--num_relax {None,Top1,Top5,All}
num_relax is 'None' (default), 'Top1', 'Top5' or
'All'. Specify how many of the top ranked structures
to relax.
--save_images save images such as structures and plot
--show_images show images
When you run the below command, totally 400 structures are generated by recycling 10 times(all structures while recycling will be output), using 10 models(5 original models and 5 pTM models), with 2 random seeds, and 2 ensemble patterns (1 and 8).
colabfold-conda/bin/python3.7 runner_af2advanced.py \
-i hoge.fasta \
-o output_dir/hoge \
--max_recycles 10 \
--num_samples 2 \
--model both \
--num_CASP14_models 5 \
--num_pTM_models 5 \
--noranking \
--output_all_cycle \
--num_ensembles 1 8If you have precomputed MSA, add --msa_method precomputed --precomputed msa.pickle.
model_1_ptm_seed_0_rec_1_ens_1.pdb # pdb file
model_1_ptm_seed_0_rec_1_ens_1.pickle # Contains pLDDT, etc.
model_1_ptm_seed_0_rec_2_ens_1.pdb
model_1_ptm_seed_0_rec_2_ens_1.pickle
...
msa.pickle
scores.csv # Contains pLDDT and pTM for predicted models
msa_coverage.pngscores.csv
,Model,pLDDT,pTMscore,Tolerance,ModelName,Seed,Recycle,Ensemble,Target
0,model_1_ptm_seed_0_rec_10_ens_1,0.8671514626497182,0.754113431050468,0.4492591,model_1_ptm,0,10,1,hoge
1,model_1_ptm_seed_0_rec_10_ens_8,0.8726767323938525,0.7699067933832592,0.58676547,model_1_ptm,0,10,8,hoge
2,model_1_ptm_seed_0_rec_1_ens_1,0.876012646623183,0.7756131184815608,19.154978,model_1_ptm,0,1,1,hoge- Mirdita M, Schuetze K, Moriwaki Y, Heo L, Ovchinnikov S and Steinegger M. ColabFold - Making protein folding accessible to all. bioRxiv, doi: 10.1101/2021.08.15.456425 (2021)
- John Jumper, Richard Evans, Alexander Pritzel, et al. - Highly accurate protein structure prediction with AlphaFold. Nature, 1–11, doi: 10.1038/s41586-021-03819-2 (2021)
The AlphaFold parameters are made available for non-commercial use only, under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. You can find details at: https://creativecommons.org/licenses/by-nc/4.0/legalcode