AI-Driven De Novo Design and Development of Nontoxic DYRK1A Inhibitors

Dual-specificity tyrosine-phosphorylation-regulated kinase 1A (DYRK1A) is implicated in several human diseases, including DYRK1A syndrome, cancer, and neurodegenerative disorders such as Alzheimer’s disease, making it a relevant therapeutic target. In this study, we combine artificial intelligence with traditional drug discovery methods to design nontoxic DYRK1A inhibitors. An ensemble QSAR model was used to predict binding affinities, while a directed message passing neural network evaluated toxicity. Novel compounds were generated using a hierarchical graph-based generative model and subsequently refined through molecular docking, chemical synthesis, and experimental validation. This pipeline led to the identification of pyrazolyl-1H-pyrrolo[2,3-b]pyridine 1 as a potent inhibitor, from which a new derivative series was developed. Enzymatic assays confirmed nanomolar DYRK1A inhibition, and additional assays demonstrated antioxidant and anti-inflammatory properties. Overall, the resulting compounds exhibit strong DYRK1A inhibition and favorable pharmacological profiles.

双特异性酪氨酸磷酸化调节激酶1A（Dual-specificity tyrosine-phosphorylation-regulated kinase 1A，DYRK1A）与多种人类疾病密切相关，包括DYRK1A综合征、癌症及阿尔茨海默病等神经退行性疾病，故而成为极具研究价值的治疗靶点。本研究将人工智能与传统药物发现手段相结合，旨在开发无毒的DYRK1A抑制剂。研究采用集成定量构效关系（quantitative structure-activity relationship, QSAR）模型预测化合物结合亲和力，同时借助定向消息传递神经网络（directed message passing neural network）评估其毒性。通过基于层级图的生成模型生成新型化合物，并经分子对接（molecular docking）、化学合成与实验验证完成后续优化。该研发流程最终筛选得到吡唑基-1H-吡咯并[2,3-b]吡啶1作为强效抑制剂，并由此开发出全新的衍生物系列。酶学实验证实其具备纳摩尔级的DYRK1A抑制活性，额外实验结果表明此类化合物同时拥有抗氧化与抗炎特性。综上，所得化合物展现出优异的DYRK1A抑制活性与良好的药理特征。