pLM-BLAST is a sensitive remote homology detection tool that is based on the comparison of residue embeddings obtained from the protein language model ProtTrans5. It is available as a standalone package as well as an easy-to-use web server within the MPI Bioinformatics Toolkit, where several precomputed databases (e.g., ECOD, InterPro, and PDB) can be searched.
Note: the method is being actively developed, please expect soon new features, such as speed improvement, global alignment, and others.
For the local use, use the requirements.txt file or environment.yml to create a new conda environment.
pip
pip install -r requirements.txt
conda
conda env create -f environment.yml
Alternatively, the packages listed below can be installed manually:
python==3.9
pandas
pytorch
scikit-learn
biopython
tqdm
numba
transformers
sentencepiece
matplotlib
Pre-calculated databases can be downloaded from http://ftp.tuebingen.mpg.de/pub/protevo/toolkit/databases/plmblast_dbs. To create a custom database, use embeddings.py script:
embeddings.py -embedder pt -cname column_name database.csv database.pt_emb.p --gpu
database.csv is an index file defining sequences and their descriptions. For example, the first lines of the ECOD database index are:
,id,description,sequence
0,ECOD_000151743_e4aybQ1,"ECOD_000151743_e4aybQ1 | 4146.1.1.2 | 4AYB Q:33-82 | A: alpha bundles, X: NO_X_NAME, H: NO_H_NAME, T: YqgQ-like, F: RNA_pol_Rpo13 | Protein: DNA-DIRECTED RNA POLYMERASE",FPKLSIQDIELLMKNTEIWDNLLNGKISVDEAKRLFEDNYKDYEKRDSRR
1,ECOD_000399743_e3nmdE1,"ECOD_000399743_e3nmdE1 | 5027.1.1.3 | 3NMD E:3-53 | A: extended segments, X: NO_X_NAME, H: NO_H_NAME, T: Preprotein translocase SecE subunit, F: DD_cGKI-beta | Protein: cGMP Dependent PRotein Kinase",LRDLQYALQEKIEELRQRDALIDELELELDQKDELIQMLQNELDKYRSVI
2,ECOD_002164660_e6atuF1,"ECOD_002164660_e6atuF1 | 927.1.1.1 | 6ATU F:8-57 | A: few secondary structure elements, X: NO_X_NAME, H: NO_H_NAME, T: Elafin-like, F: WAP | Protein: Elafin",PVSTKPGSCPIILIRCAMLNPPNRCLKDTDCPGIKKCCEGSCGMACFVPQ
Index can be generated from a FASTA file using scripts/makeindex.py.
Use -cname to specify in which column of the database.csv file sequences are stored
The resulting embeddings will be stored in database.pt_emb.p
Usage of --gpu is highly recommended (cpu calculations are orders of magnitude slower)
To search a pre-calculated or custom database, follow scripts/example.sh
import torch
import alntools as aln
from alntools.base import Extractor
import pandas as pd
# load the embedding; for the embedding calculation, refer to `scripts/example.sh`
emb_file = './scripts/output/A9A4Y8.pt_emb.p'
embs = torch.load(emb_file)
# a self-comparison will be performed
seq1_emb, seq2_emb = embs[0], embs[0]
# calculate embedding similarity aka substitution matrix
densitymap = aln.density.embedding_similarity(seq1_emb, seq2_emb)
# convert to numpy array
densitymap = densitymap.cpu().numpy()
# find all alignment possible paths (traceback from borders)
paths = aln.alignment.gather_all_paths(densitymap)
# score those paths
results = aln.prepare.search_paths(densitymap, paths=paths, as_df=True)
# remove redundant hits
results = aln.postprocess.filter_result_dataframe(results)
If you find the pLM-BLAST useful, please cite the preprint:
"pLM-BLAST – distant homology detection based on direct comparison of sequence representations from protein language models"
Kamil Kaminski, Jan Ludwiczak, Vikram Alva, and Stanislaw Dunin-Horkawicz
bioRxiv https://www.biorxiv.org/content/10.1101/2022.11.24.517862v1
If you have any questions, problems or suggestions, please contact us.
This work was supported by the First TEAM program of the Foundation for Polish Science co-financed by the European Union under the European Regional Development Fund.