Unsupervised clustering algorithm for identifying and classifying protein domains based on structural similarity.
DPCstruct uses the moodycamel::ConcurrentQueue library freely available provided citation (Simplified BSD license).
- GNU compiler (version 13.0 or higher)
- CMake build system
- Foldseek (version 9 or higher)
To install DPCstruct, clone the repository and run the following commands:
git clone https://github.com/RitAreaSciencePark/DPCstruct.git
cd DPCstruct
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=/path/to/install ..
make -j 4
make install
DPCstruct is designed to process local structural alignments produced by Foldseek, enabling comprehensive identification and classification of protein domains based on structural similarity. This guide explains each module in the pipeline and how it contributes to the overall process.
To ensure compatibility with DPCstruct, please verify that your local alignments have the correct format. For detailed instructions on how to generate DPCstruct-compatible all-vs-all alignments, please refer to the How to generate all-vs-all alignments section.
The pipeline is organized into five main modules. Each module has dedicated commands and help documentation for a smooth execution:
dpcstruct <module> -h
Modules:
Step 1) prefilters: applies a series of filters to the alignments found with Foldseek.
Step 2) primarycluster: clusters local alignments per query sequence.
Step 3) secondarycluster: clusters the primary clusters.
Step 4) traceback: traces back the alignments to the original sequences.
Step 5) postfilters: removes redundancies from the secondary clusters.
The final output is a TSV file which includes the following columns:
- protIndex: Protein identifier.
- dom-start: Starting index of the domain.
- dom-end: Ending index of the domain.
- metaclusterID: Identifier for the assigned structural metacluster.
The folder example contains a toy example to test the pipeline.
To run the example you can simply execute run_example.sh.
| File | Description |
|---|---|
| proteins.tsv | list of proteins and their corresponding index |
| alns.zip | set of all-vs-all local alignments from where to start |
In the following section we provide a quick guide on how to generate local alignments as input to DPCstruct.
Given a folder containing a set of pdbs [*.pdb], the standard procedure to generate the local alignments is the following. For more information check Foldseek repo.
# generate protein index table
ls ${pbsDir}/*.pdb | awk '{print NR,$1}' > ${fsdbDir}/${proteinLookup}
# change pdbs filenames to indexes (this step can be performed after all vs. all)
while IFS=' ' read -r index file; do mv "${file}.pdb" "${index}"; done < ${fsdbDir}/${proteinLookup}
# create database
foldseek createdb ${pdbsDir}$/ ${fsdbDir}/pdbs_db
# run local alns (adjust e-value as required)
foldseek search ${queryDB} ${targetDB} ${alns} ${tmpDir} -a --threads ${SLURM_CPUS_PER_TASK}
foldseek convertalis ${queryDB} ${targetDB} ${alns} ${alnsConverted} --format-mode 4 --format-output query,target,qstart,qend,tstart,tend,qlen,tlen,alnlen,pident,evalue,bits,alntmscore,lddt
If you are using protein structure predictions from AlphaFold, you will need to download the per-residue pLDDT values for each protein in your dataset.
In the ./build/util/ folder, we provide an auxiliary script download_plddts.sh to download this information from the AlphaFold Database hosted on Google Cloud Public Datasets and store it in a binary format compatible with the prefiltering module.