Chorus is an efficient protein-protein local alignment search tool for multiple query sequences and large databases. Our method is up to 300x faster than NCBI-BLASTP when inputting multiple query seqs at a time, and can stay ahead of the state-of-the-art database-index methods when there are less than 5000 input query seqs (2.2-27x faster than fastest MMseqs2, 1.1-9x than fastest DIAMOND), while maintaining a low memory footprint.
Rapid multiple protein sequence search by parallel and heterogeneous computation
Hardware:
A linux (ubuntu) server with a NVIDIA GPU.
Software:
g++ 9
CUDA 11.7
CUDA Environment (add to ~/.bashrc):
export CUDA_HOME=/usr/local/cuda
export PATH=$CUDA_HOME/bin:$PATH
export CPATH=$CUDA_HOME/include:$CPATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH
*Don't forget to source ~/.bashrc
mkdir build
cd build
cmake ..
make
2.1. Download and build tantan: https://gitlab.com/mcfrith/tantan
2.2. Tantan the database
tantan -p -x X <protein_db.fasta> > <protein_db_tantaned.fasta>
2.3. Tantan the query sequences
tantan -p -x X <query.fasta> > <query_tantaned.fasta>
Input: A protein database file (in fasta format).
./createDB <protein_db(_tantaned).fasta> <output db> [batch size (GB)]
Output: This step will create database files for each part, including "dbnameX.seq", "dbnameX.name", "dbnameX.sofs" and "dbnameX.nofs", to the same directory with the original db file.
If the batch size not specified, the default size is one-third of the GPU memory.
E.g.
./createDB ../db/nr.fasta ../db/nr 4
Input: The pre-processed database in step 2. And a query sequences file (in fasta format).
./query -q <query(_tantaned).fasta> -d <db1> [db2 ... (optional)] -o <output file>
Chorus will automatically detect all parts of the database created in the same directory, iteratively process them, and finally show all results in the output file.
E.g.
./query -d ../db/nr -q ../example_query -o res.out
Please use
./query --help
to get help message.
| Arg | Description | Default |
|---|---|---|
| --outfmt | Output format: 0: m8 tabular (like blast+ outfmt 6); 1: detailed alignment; 2: tabular and ref seqs (fasta); 3: Opfi format; 4: a3m format; 5: Diamond benchmark format. | 0 |
| -p, --filter-level | The filter level to pass a database sequence to smith waterman align. The smaller the value, the more sensitive the result. ("sensitivity" refers to the ability of a sequence alignment algorithm to correctly identify and align similar or homologous sequences, especially those that share low levels of similarity or are distantly related.) | 1 |
| --min-score | The minimum score for each alignment to display. | 0 |
| -e, --max-evalue | The maximum expectation value for each alignment to display | 1e1 |
| --max-output-align | The maximum number of alignments to display for each query. (0 means no limit) | 0 |
| --num-threads | Number of CPU threads. | Maximum threads available |
| -k, --seed-length | Length of k-mer. (3, 4, or 5) | 5 |
| -w, --qit-width | Width of query index table. (Number of indices for each row) The larger the value, the less likely the overflow occurs. | 4 |
| -h, --hash-size | Voting hash table size ratio. The larger the value, the larger the GPU voting hash table (doubly for each increment). (0: maybe OK; 1: OK for most query seqs; 2: redundancy) | 2 |
| --band-width | The bandwidth in banded smith-waterman. The larger the value, the more sensitive the result. | 8 |
| --must-include | Requires the ref sequence must match a regular expression n times. (e.g. --must-include "R[A-Z]{4,7}H" 2) | No limit |
We offer Docker services for Chorus. For more information, please visit our Docker Hub page at bioacc/chorus.
Manuscript Reproduction Scripts: "Rapid Multiple Protein Sequence Search by Parallel and Heterogeneous Computation"
If you want to reproduce the experimental results in the manuscript. please visit: Chorus Reproduction Scripts.