Abstract: Drug-target association prediction is researched. PDML based on weak label learning and multi-information fusion is proposed to find new drug-target interactions from human enzymes, ion channels, GPCRs and nuclear receptors. The performance of the proposed method makes better than the methods provided by Yamanishi, RLSMDA, LapRLS and NetCBP in terms of sensitivity, specificity, AUC values and AUPR values except that the AUC values of the model slightly decrease in nuclear receptor dataset compared to LapRLS. The five drug-target interaction pairs with highest scores can be extracted and validated by available public database DrugBank, SuperTarget and KEGG.

Key words: drug-target interaction, multi-label learning, multi-information fusion, drug-target interaction network, drug similarity

摘要： 对药物-靶标关联进行了研究，提出基于弱标记和多信息融合的药物-靶标相互作用预测方法PDML。通过与其他方法对比和数据库检索验证评估PDML模型的性能：与Yamanishi提出的方法、RLSMDA、LapRLS及NetCBP相比，除在核受体数据集中该方法在AUC上的性能比LapRLS略有降低之外，模型在敏感性、特异性、AUC和AUPR上的性能均优于其他四种方法；提取前5个预测分值最高的药物-靶标对，这些药物-靶标对能通过检索DrugBank、SuperTarget和KEGG数据库而得到验证。

关键词: 药物-靶标相互作用, 多标记学习, 多信息融合, 药物-靶标相互作用网络, 药物相似性