hugging-science/jain-clinical-antibody-polyreactivity

该数据集包含86个临床阶段抗体重链可变区（VH）序列及其二元多反应性标签，用于重现Sakhnini等人（2025年）的基准测试结果。原始数据由Jain等人（2017年）发布，包含137种FDA批准或晚期临床抗体的生物物理测量。数据集用于评估基于Boughter数据集训练的ESM-1v + Logistic Regression模型，实现了与Novo Nordisk公布结果的完全一致。关键特征包括：人类临床阶段抗体、抗体重链可变区序列、ELISA多反应性面板（6种配体）、二元分类标签（0=特异性，1=非特异性/多反应性）、86个抗体样本（57个特异性，29个非特异性）。数据集支持二元分类任务和Novo Nordisk基准测试，准确率为68.60%。

This dataset contains 86 clinical-stage antibody heavy chain variable domain (VH) sequences with binary polyreactivity labels, preprocessed to reproduce the benchmark results from Sakhnini et al. (2025) (Novo Nordisk & University of Cambridge). The original dataset was published by Jain et al. (2017) and contains biophysical measurements for 137 FDA-approved or late-stage clinical antibodies. This is a benchmark test set for evaluating the ESM-1v + Logistic Regression model trained on the Boughter dataset, achieving exact parity with Novos published results (68.60% accuracy). Key features include: human clinical-stage antibodies, antibody VH sequences, ELISA polyreactivity panel (6 ligands), binary classification labels (0 = specific, 1 = non-specific/polyreactive), and 86 antibodies (57 specific, 29 non-specific). The dataset supports binary classification tasks and the Novo Nordisk parity benchmark.