基于分子描述符和机器学习方法预测和虚拟筛选乳腺癌靶向蛋白HEC1抑制剂

摘要

Highly expressed in cancer 1 (HEC1) is a conserved mitotic regulator that is critical for spindle checkpoint control, kinetochore functionality, and cel survival. Overexpression of HEC1 has been detected in a variety of human cancers, and it is linked to poor prognosis of primary breast cancers. Thus, it is important to screen novel inhibitors with high affinity for HEC1. Machine learning (ML) methods were exhibiting good predicting capability in several aspects of the diverse compounds, such as pharmacokinetics, pharmacodynamics, and toxicity. In this work, two ML methods, support vector machines (SVMs) and random forests (RFs), were used to develop a classification method for searching inhibitors and non-inhibitors of HEC1 from the chemical library of structural diversity by screening characteristics of molecular descriptors. Both ML methods achieved promising prediction accuracies, and the RF model showed better performance. We performed virtual screening of HEC1 inhibitors by the RF model from an in-house database to screen potential HEC1 inhibitors. Two novel potential candidates were found.In vitro experiments of the two compounds showed that both had a certain degree of antitumor activity for the MDA-MB-468 and MDA-MB-231 breast cancer cel lines. Our study shows that ML methods are promising to design and virtualy screen inhibitors of HEC1.%HEC1(癌症高表达蛋白)是纺锤体检查点控制、着丝粒功能、细胞存活的关键的有丝分裂调节器，与原发性乳腺癌的不良预后有关。筛选具有高亲和力的HEC1新型抑制剂对探索乳腺癌的靶向治疗具有重要意义。本文从结构多样性的化合物库中筛选HEC1抑制剂。通过对分子描述符的特征筛选，采用支持向量机(SVM)和随机森林(RF)方法分别对HEC1抑制剂和非抑制剂建立了分类模型。经对比， RF模型显示了更好的预测精度。我们采用RF模型对HEC1抑制剂进行了虚拟筛选，从“in-house”实体库筛选得到2个潜在的HEC1抑制剂分子。随后对筛出的化合物进行了体外活性实验，发现对乳腺癌细胞株MDA-MB-468和MDA-MB-231均有一定程度的抗肿瘤活性。研究结果表明，机器学习方法对于设计和虚拟筛选HEC1抑制剂有良好的效果。