PCAC-Affinitydata

This dataset contains antigen-antibody binding affinity data compiled from multiple established databases for machine learning applications in immunology research. The dataset includes sequence-based information from five major sources: PaddlePaddle 2021 Antibody Dataset, AB-Bind, SKEMPI 2.0, SAbDab (Structural Antibody Database), and a derived Benchmark dataset.The compiled dataset comprises a total of 18,651 raw entries, which were filtered and preprocessed to yield 6,956 high-quality data points. The PaddlePaddle 2021 dataset serves as the primary training and validation set with 4,875 entries (originally 5,210), containing AG and Kd-derived entries with explicit chain sequences. Additional datasets include AB-Bind (691 entries from 1,101 raw), SKEMPI 2.0 (387 entries from 7,085 raw), SAbDab (579 entries from 3,241 raw), and Benchmark (264 entries from 2,014 raw).All data are provided in sequence format, with preprocessing steps applied to retain only wild-type antigen-antibody complexes with available chain sequences and measured binding affinities. The dataset follows a 6:2:2 training-validation-testing split ratio for the main datasets, while the Benchmark dataset serves as an independent test set to ensure robust model evaluation with less than 30% sequence identity to training antigens.This curated dataset enables researchers to develop and evaluate machine learning models for predicting antigen-antibody binding affinity, supporting computational immunology and therapeutic antibody design applications.