srescalli/X-PAIR_datasets

X-PAIR datasets

This repository contains the processed datasets used to train and evaluate X-PAIR, an ultrafast multitask framework for protein–protein interaction (PPI) and partner-specific interface prediction from protein sequences.

The datasets are provided in the train, validation, and test splits used in the experiments reported in the X-PAIR study.

Datasets

The repository contains two groups of datasets:

1. Datasets generated in the X-PAIR study
2. Processed versions of previously published benchmark datasets

X-PAIR datasets

The following datasets were generated and used in this study:

Dataset	Description
X-fair	Dataset designed for joint PPI and partner-specific interface prediction, with homology-aware splitting to reduce information leakage.
X-human	Human training and validation dataset designed to evaluate cross-species generalization.
X-multispecies	Multispecies training and validation dataset comprising human, chicken, fly, and yeast data, designed to evaluate cross-species generalization.
X-taxonomic-interface	Taxon-specific interface datasets designed to investigate the evolutionary generalization of interface prediction.

Protein–protein interaction data used to construct these datasets were obtained from STRING v12, while structural protein–protein interface annotations were obtained from PPI3D.

The X-PAIR datasets generated in this study are also archived on Zenodo.

External benchmark datasets

Three previously published datasets were used as external benchmarks. For computational feasibility, the original datasets were processed by excluding protein pairs containing sequences longer than 2,000 amino acids from the training and validation sets, while the original test sets were retained without filtering. This repository provides the resulting processed train, validation, and test splits exactly as used to train and evaluate X-PAIR.

Benchmark	Task	Original source
Cross-species benchmark (Sledzieski et al.)	PPI prediction	https://zenodo.org/records/5140612
Gold Standard (Bernett et al.)	PPI prediction	https://doi.org/10.6084/m9.figshare.21591618.v3
PIONEER (Xiong et al.)	Partner-specific interface prediction	https://static-content.springer.com/esm/art%3A10.1038%2Fs41587-024-02428-4/MediaObjects/41587_2024_2428_MOESM4_ESM.xlsx

Model

X-PAIR can be installed from PyPI:

pip install xpair

The complete X-PAIR source code is publicly available on GitLab.

Installation and usage instructions are provided in the README, while detailed software documentation is available in the X-PAIR documentation.

Trained model checkpoints are available on Hugging Face.

Citation

If you use X-PAIR or the datasets generated in this study, please cite:

Rescalli, S. & Carbone, A.
X-PAIR: an ultrafast multitask framework for proteome-scale reconstruction of PPI networks and partner-specific interfaces from sequence.
bioRxiv (2026).
https://doi.org/10.64898/2026.07.20.739596

When using one of the external benchmark datasets, please also cite the corresponding original publication.