Code supporting the Protein Design Archive (PDA) database publication
Welcome to the Protein Design Archive (PDA) Database, an up-to-date, complete, online resource for the protein design community and beyond. Here is a description of what can be found in each directory.
Due to the nature of the PDA and resources that it builds on, we believe that presenting high-quality, accurate and complete information requires manual curation of the dataset. In order to make this process transparent, we use for the curation version-controlled files and scripts kept in this repository. These are:
entries_to_manually_exclude.csvwhich lists PDB codes that have come up in search for designed proteins on the RCSB PDB, but which we believe are not relevant to the field of de novo protein design.entries_to_manually_include.csvwhich lists PDB codes that we believe are relevant to the field of de novo protein design, but do not come up in search for designed proteins due to their labelling (predominantly the "older" entries).entries_manually_reviewed.csvwhich lists PDB codes that we have manually reviewed and confirm our belief that they belong to the PDA - this will grow with time and eventually contain the whole dataset.
Edits and comments to the data entries themselves can be found in scripts/manual_data_curation.py.
Edit: code for data preparation used for the monthly updates of the PDA has been much improved since October 2024. In view of much interest in this resource and commitment to maintaining it to the highest standard, updates since October 2024 are performed using the improved method, with accurate designed chain selection and a semi-automated method, reducing chance of inconsistency and errors that could potentially arise
Data is now collected using the main.py script, run with flags --next iso_format_date_of_next(current)_update --prev iso_format_date_of_previous_update e.g. --next 20241031 --prev 20240930 for the October 2024 update which includes entries up to 31st October 2024.
This script relies on other scripts found in scripts directory, but should be self explanatory with the comments and cues printed to the command line.
Below is described the original method used at the time of submission of manuscript "The Protein Design Archive (PDA): insights from 40 years of protein design" (doi: https://doi.org/10.1101/2024.09.05.611465) on 7th September 2024. This method has been improved since them (see section above).
Data has been scraped from the RCSB PDB database using the Jupyter notebook Prepare_PDA_data.ipynb.
It relies on having a list of PDB codes to scrape data for, cif files to scrape information from, and in-house rules for assigning suggested classification for the protein designs.
This base dataset is supplemented using chain labels from each chain's description in FASTA file and Extract_auth_chain_labels.ipynb notebook.
To download PDB files for Foldseek and DE-STRESS analysis, run the following commands:
chmod +x download_pdbs.sh
bash download_pdbs.sh -f pdb_codes.csv -o ./output/pdb_files -a
Analysis of similarity measures relies on checking whether the target that the query has matched with has an earlier release date than the query. To do this, all_pdb_release_dates.csv file is used, which is prepared with the Dates_of_release_of_all_PDB.ipynb Jupyter notebook.
The most up-to-date dataset of designed proteins can be found in the file 20240827_data.json.
Jupyter notebook Concatenate_chain_sequences.ipynb contains the code used to extract only designed chains from original FASTA files. designed_single_sequences_1450.fasta is the single designed chains input for analysis. DvD and DvP directories each contain chain and struct directories, to reflect their type of searches (Designs vs Designs, or Designs vs Natural Proteins; per single designed chains or per concatenated designed chains). Each directory contains the script used to run the software analysis (search_{dvd-or-dvp}.sh), and .json outputs of the highest scores and highest scoring partners for every design, and every metric. .json analysis outputs were prepared using Similarity_analysis.ipynb Jupyter notebook, which can be found within MMseqs2 directory, and which also provides the overall similarity analysis and graph plotting.
Jupyter notebook Extract_designed_chains_in_pdb_files.ipynb contains the code used to extract only designed chains from original PDB files. DvD and DvP directories each contain chain and struct directories, to reflect their type of searches (Designs vs Designs, or Designs vs Natural Proteins; per single designed chains or per concatenated designed chains). Each directory contains the script used to run the software analysis (foldseek_search_{dvd-or-dvp}_{single_chain-or-concatenated}.sh) a python file to analyse the results and extract the highest scoring partner and value for each metric (e.g. foldseek_all_metrics_above_threshold.py), and .json outputs of the highest scores and highest scoring partners for every design, and every metric.
The de-stress-analysis/ folder contains all the code needed to produce the amino acid/secondary structure composition heatmaps and box plots of the de-stress metrics that are included in figure 4.
First, create the conda environment and activate it with the command below.
conda env create -f de-stress-analysis/environment.yml
conda activate pda_destress_analysisAfter this download destress_data_designs_082024.csv, destress_data_pdb_082024.csv, design_meta_data_20240827.json, designed_chains_pdb_files/ and pdb_all_files_june_2024/ from the supplementary materials and place them in the de-stress-analysis/data/raw_data/ folder.
Next, run the de-stress-analysis/data_prep_destress.py and de-stress-analysis/data_prep_aa_comp.py scripts to prepare the data.
Finally, run the de-stress-analysis/analysis_destress.py and de-stress-analysis/analysis_aa_comp.py scripts to produce the plots that were used in the paper and these will be saved in the de-stress-analysis/analysis/ folder.
Code for plotting protein designs' growth curve.