物理化学学报 >> 2014, Vol. 30 >> Issue (1): 171-182.doi: 10.3866/PKU.WHXB201311041

生物物理化学 上一篇    下一篇

基于分子描述符和机器学习方法预测和虚拟筛选MMP-13对MMP-1的选择性抑制剂

李秉轲1, 丛湧1, 田之悦1, 薛英1,2   

  1. 1 四川大学化学学院, 教育部绿色化学与技术重点实验室, 成都 610064;
    2 四川大学生物治疗国家重点实验室, 成都 610041
  • 收稿日期:2013-07-05 修回日期:2013-11-04 发布日期:2014-01-01
  • 通讯作者: 薛英 E-mail:yxue@scu.edu.cn
  • 基金资助:

    国家自然科学基金(21173151)资助项目

Predicting and Virtually Screening the Selective Inhibitors of MMP-13 over MMP-1 by Molecular Descriptors and Machine Learning Methods

LI Bing-Ke1, CONG Yong1, TIAN Zhi-Yue1, XUE Ying1,2   

  1. 1 College of Chemistry, Key Laboratory of Green Chemistry and Technology, Ministry of Education, Sichuan University, Chengdu 610064, P. R. China;
    2 State Key Laboratory of Biotherapy, Sichuan University, Chengdu 610041, P. R. China
  • Received:2013-07-05 Revised:2013-11-04 Published:2014-01-01
  • Contact: XUE Ying E-mail:yxue@scu.edu.cn
  • Supported by:

    The project was supported by the National Natural Science Foundation of China (21173151).

PDF

1947

赞

摘要：

基质金属蛋白酶-13 (MMP-13)为预防和治疗骨关节炎(OA)提供了充满希望的靶标. 通过抑制剂来阻断MMP-13的活性将会对治疗OA疾病产生潜在的作用. 然而，宽谱抑制剂同样抑制MMP家族的其它成员，特别是MMP-1，这将会导致肌与骨的综合症. 因此，设计和发现潜在的MMP-13 相对于MMP-1 的高效选择性抑制剂，在对治疗OA新型药物的研发中具有相当重要的现实意义. 本研究通过两种机器学习方法(ML)：支持向量机(SVM)和随机森林(RF)来建立分类模型，用于预测不同结构的MMP-13 对MMP-1 的选择性抑制剂. 所建这些模型的预测效果都已经达到了令人满意的精度. 在这两种ML模型中，RF对于MMP-13选择性抑制剂和非抑制剂的精度分别达到97.58%和100%. 同时，与MMP-13对MMP-1的选择性抑制最相关的分子描述符也基于不同的特征选择方法被两种模型挑选出来. 最后，用预测效果最好的RF模型虚拟筛选了ZINC数据库的“fragment-like”子集，从而得到了一系列潜在的候选药物. 研究表明，机器学习方法，特别是RF方法，对于发现潜在的MMP-13选择性抑制剂十分有效. 同时还得到了一些与MMP-13的选择性抑制相关的分子描述符.

关键词: 基质金属蛋白酶-13, 选择性抑制剂, 机器学习方法, 支持向量机, 随机森林, 虚拟筛选

Abstract:

Matrix metalloproteinase-13 (MMP-13) is an interesting target for the prevention and therapy of osteoarthritis (OA). Interruption of MMP-13 activity with an inhibitor has the potential to affect OA. However, a broad-spectrum inhibitor, which restrains the other members of the MMP family, especially MMP-1, can cause musculoskeletal syndrome. So, the design and discovery of potential and highly selective inhibitors for MMP-13 over MMP-1 are necessary and of great significance for the development of novel therapeutic agents against OA. Two machine-learning (ML) methods, support vector machine and random forest (RF), were explored in this work to develop classification models for predicting selective inhibitors of MMP-13 over MMP-1 from diverse compounds. These ML models achieved promising prediction accuracies. Among the two ML models, RF gave the better performance, i.e., 97.58% for MMP-13 selective inhibitors and 100% for non-inhibitors. We also used different feature selection methods to extract the molecular features most relevant to selective inhibition of MMP-13 over MMP-1 from the two models. In addition, the betterperforming RF model was used to perform virtual screening of MMP-13 selective inhibitors against the "fragment-like" subset of the ZINC database to enrich the potential active agents, thereby obtaining a series of the most potent candidates. Our study suggests that ML methods, particularly RF, are potentially useful for facilitating the discovery of MMP-13 inhibitors and for identifying the molecular descriptors associated with MMP-13 selective inhibitors.

Key words: Matrix metalloproteinase-13, Selective inhibitor, Machine learning method, Support vector machine, Random forest, Virtual screening