Data associated with “Biophysical prediction of protein–peptide interactions and signaling networks using machine learning”

Data associated with “Biophysical prediction of protein–peptide interactions and signaling networks using machine learning” by Joseph M. Cunningham, Julia R. Rogers, Grigoriy Koytiger, Peter K. Sorger, and Mohammed AlQuraishi (https://doi.org/10.1038/s41592-019-0687-1). General descriptions are provided below with further details given in READMEs. <strong>HSM data</strong> All data used for training and evaluating HSM is contained in <em>data.zip</em>. This includes pre-processed domain-level data; raw (unaligned) domain-level data; publication sources for HSM training data; pre-processed domain and peptide metadata used in HSM/P proteome-level predictions; and experimental protein-level data used to evaluate HSM/P. <strong>HSM training results</strong> Complete results from hyperparameter tuning and cross-fold evaluation of HSM/ID and HSM/D are contained in <em>training.zip</em>. <strong>PBD model comparisons</strong> Results from PSSMs and external models (PepInt and NetPhorest) evaluated on HSM domain-level data are contained in <em>comparison_models.zip</em>. <strong>HSM/P predictions</strong> Complete set of HSM/P predictions for ‘gold standard’ dataset and human proteome are contained in <em>hsmp_predictions.zip</em>. <strong>Published results</strong> Results from all analysis described in the publication along with source data for figures are contained in <em>publication_results.zip</em>. This also includes domain sequences and multiple sequence alignments used in both HSM/D and HSM/P and a complete list of novel HSM/P predicted PPIs in the human proteome (<em>results/data/perf/hsmp/ppi_predictions.fdr0.01.csv</em>).