Awesome
AF-Cluster
Code and data corresponding to Wayment-Steele*, Ojoawo*, ... Ovchinnikov, Colwell, Kern (2023) "Predicting multiple conformations via sequence clustering with AlphaFold2" Nature. link
[Aug 2024] AF-Cluster can be combined with random seeds and MSA subsampling in this Colab Notebook
[Jan 2024] An exact set of methods to reproduce every structure prediction in the paper can be found here.
[Jan 2023] (DEP, use ColabDesign based notebook above). Run the entirety of the script in a Colab Notebook here!
Usage
To generate MSA:
All MSAs used in this manuscript were generated using the ColabFold notebook.
To cluster MSA and generate subsampled MSA files:
python scripts/ClusterMSA.py EX -i initial_msa.a3m -o msas
Outputs a directory named msas that contains
- msas/EX_000.a3m
- msas/EX_001.a3m
...
- msas/EX_REF.a3m
- msas/EX_U10-000.a3m
- msas/EX_U10-001.a3m
...
- msas/EX_U100-000.a3m
- msas/EX_U100-001.a3m
...
- msas/EX_REF.a3m
- msas/EX_clustering_assignments.tsv
-msas/EX_cluster_metadata.tsv
EX_000.a3m, EX_001.a3m ... are the clusters identified by DBSCAN.
EX_U10-000.a3m, ... EX_U10-009.a3m are uniformly sampled control MSAs of size 10 (Default is to generate 10).
EX_U100-000.a3m, ... EX_U100-009.a3m are uniformly sampled control MSAs of size 100 (Default is to generate 10).
EX_REF.a3m is a copy of the original MSA.
EX_clustering_assignments.tsv contains a list of original sequences and the cluster index they were assigned to (-1 means they were not assigned).
EX_cluster_metadata.tsv contains metadata corresponding to clusters.
To also perform PCA and/or tSNE embedding at the same time and save it in EX_clustering_assignments.tsv for later analysis:
python scripts/ClusterMSA.py -i <my_alignment.a3m> -o <outdir> --run_PCA
or
python scripts/ClusterMSA.py -i <my_alignment.a3m> -o <outdir> --run_tSNE
Example output for KaiB:
2OUG
2006 seqs removed for too many gaps, 4745 remaining
eps n_clusters n_not_clustered
3.00 1 1186
4.00 1 1186
5.00 3 1180
6.00 8 1161
7.00 12 1147
8.00 36 1045
9.00 39 950
10.00 62 825
Selected eps=10.00
4745 total seqs
315 clusters, 2280 of 4745 not clustered (0.48)
avg identity to query of unclustered: 0.30
avg identity to query of clustered: 0.37
wrote clustering data to msas/2OUG_clustering_assignments.tsv
wrote cluster metadata to msas/2OUG_cluster_metadata.tsv
writing 10 size-10 uniformly sampled clusters
writing 10 size-100 uniformly sampled clusters
To run AF2:
python scripts/RunAF2.py
See https://github.com/jproney/AF2Rank for more information on compiling an AlphaFold2 installation.
To run MSA Transformer:
python scripts/runESM.py -i <my_subMSA.a3m> -o <outdir>
To calculate RMSD to provided reference structure(s):
python scripts/CalculateModelFeatures.py path/to/pdbs/* -o <my_output_file>.json.zip --ref_struct REF_PDB_1.pdb REF_PDB_2.pdb
To reproduce figures in preprint:
See .ipynb files included in relevant folders in data.