AI-enhanced computational discovery of promising ALK5 inhibitors in a ultra-large chemical space library for cardiovascular Disease therapy

Cardiac fibrosis, characterized by excessive extracellular matrix deposition, is a critical contributor to cardiovascular diseases, including heart failure. Transforming growth factor-beta 1 signaling, especially through activin receptor-like kinase 5 (ALK5), plays a key role in cardiac fibroblast activation and fibrosis. Traditional drug discovery approaches face challenges in identifying ALK5 inhibitors. This study leverages computational methods to expedite the discovery of potential ALK5 inhibitors. An active learning model was trained to screen a vast compound library, resulting in the selection of promising candidates. Molecular fingerprint clustering analysis and the absorption, distribution, metabolism, excretion, toxicity evaluation further characterized these compounds. Machine learning-based quantitative structure - activity relationship models predicted their activity. Molecular dynamics simulations assessed binding stability in different environments. DE50349483 and DE21377883 emerged as promising candidates with potential inhibitory effects. This study showcases the power of computational methods in drug discovery, offering hope for innovative therapies in cardiac fibrosis.

心肌纤维化（Cardiac fibrosis）以细胞外基质过度沉积为典型特征，是包括心力衰竭在内的心血管疾病的关键致病因素。转化生长因子-β1（Transforming growth factor-beta 1）信号通路，尤其是通过激活素受体样激酶5（ALK5）介导的信号转导，在心肌成纤维细胞活化及纤维化进程中发挥关键作用。传统药物研发方法在筛选ALK5抑制剂方面面临诸多挑战。本研究借助计算方法加速潜在ALK5抑制剂的发现进程：训练主动学习模型（active learning model）对大规模化合物库（compound library）进行筛选，最终遴选出具有开发潜力的候选化合物。通过分子指纹（molecular fingerprint）聚类分析（clustering analysis）以及吸收、分布、代谢、排泄与毒性评价，进一步对这些候选化合物进行了系统性表征。基于机器学习（machine learning）的定量构效关系（quantitative structure-activity relationship）模型对这些化合物的生物活性进行了预测。通过分子动力学模拟（molecular dynamics simulations）评估了候选化合物与靶点在不同环境下的结合稳定性。最终，DE50349483与DE21377883被证实为极具开发潜力的候选化合物，具备潜在的抑制活性。本研究充分展现了计算方法在药物研发中的强大应用潜力，为心肌纤维化领域的创新治疗方案提供了新的希望。