NetCal-DTI: A Network Topology-Calibrated Hybrid Framework for High-Precision Inductive Drug–Target Interaction Prediction

All datasetsin benchmark test are available at our GitHub repository (https://github.com/JuFanbo/NetCal-DTI).We obtained DrugBank V6.0 and BIOSNAP from https://go.drugbank.com/releases/latestand https://snap.stanford.edu/biodata/, respectively. The subset of BindingDB with Kdvalues can be downloaded from https://www.kaggle.com/datasets/christang0002/bindingdb-for-dta. Thesingle-cell transcriptomic data for Tucidinostat-treated samples were retrieved from the public dataset GSE301188: https://www.ncbi.xyz/geo/query/acc.cgi?acc=GSE301188.Results from 5 independent runs of our ablation study on DrugBank, BIOSNAP and BindingDB are provided in Supplementary Data 1. Details for LEsimilarity heatmap are provided in Supplementary Data 2. Prediction Score Comparison Between Calibrated Model and Purely Inductive Baseline on a test set are provided in Supplementary Data 3. Top100 and Last100 docking results of NetCal-DTI predictions on CDK2 and SERT are provided in Supplementary Data 4. Target screening results for Tucidinostat on 100 runs are provided in Supplementary Data 5.本基准测试所用的全部数据集均可从我们的GitHub仓库（https://github.com/JuFanbo/NetCal-DTI）获取。其中，DrugBank V6.0与BIOSNAP数据集分别来源于https://go.drugbank.com/releases/latest与https://snap.stanford.edu/biodata/；带解离常数（Kd）值的BindingDB（结合数据库）子集可从https://www.kaggle.com/datasets/christang0002/bindingdb-for-dta下载。西达本胺（Tucidinostat）处理样本的单细胞转录组数据取自公开数据集GSE301188：https://www.ncbi.xyz/geo/query/acc.cgi?acc=GSE301188。我们在DrugBank、BIOSNAP及BindingDB上开展的5次独立消融实验结果已收录于补充数据1；LE相似性热图的详细说明见补充数据2；校准模型与纯归纳基线模型在测试集上的预测得分对比结果详见补充数据3；NetCal-DTI对CDK2与SERT的预测结果中排名前100及后100的对接结果已在补充数据4中提供；西达本胺的100次靶点筛选结果收录于补充数据5。