Kazi Md Fahmi, Haque Shahil Yasar, Jahan Eashrat, Chakma Latin, Shermin Tamanna, Uddin Ahmed Asif, Islam Salekul, Shatabda Swakkhar, Azim Riasat
Department of Computer Science and Engineering, United International University, Dhaka 1212, Bangladesh.
Department of Electrical Computer and Engineering, North South University, Dhaka 1229, Bangladesh.
Bioinform Adv. 2025 Feb 19;5(1):vbaf031. doi: 10.1093/bioadv/vbaf031. eCollection 2025.
Structure-based drug design (SBDD) holds promising potential to design ligands with high-binding affinity and rationalize their interaction with targets. By utilizing geometric knowledge of the three-dimensional (3D) structures of target binding sites, SBDD enhances the efficacy and selectivity of therapeutic agents by optimizing binding interactions at the molecular level. Here, we present CoDNet, a novel approach that combines the conditioning capabilities of ControlNet with the potency of the diffusion model to create generative frameworks for molecular compound design. This proposed method pioneers the application of ControlNet in diffusion model-based drug development. Its ability to generate drug-like compounds from 3D conformations is prominent due to its capability to bypass Open Babel post-processing and integrate bond details and molecular information.
For the gold standard QM9 dataset, CoDNet outperforms existing state-of-the-art methods with a validity rate of 99.02%. This competitive performance underscores the precision and efficacy of CoDNet's drug design, establishing it as a significant advancement with great potential for enhancing drug development initiatives.
基于结构的药物设计（SBDD）在设计具有高结合亲和力的配体并使其与靶点的相互作用合理化方面具有广阔的潜力。通过利用靶点结合位点的三维（3D）结构的几何知识，SBDD通过在分子水平上优化结合相互作用来提高治疗药物的疗效和选择性。在此，我们提出了CoDNet，这是一种将ControlNet的条件能力与扩散模型的效能相结合的新方法，用于创建分子化合物设计的生成框架。该方法开创了ControlNet在基于扩散模型的药物开发中的应用。由于其能够绕过Open Babel后处理并整合键细节和分子信息，它从3D构象生成类药物化合物的能力非常突出。
对于金标准QM9数据集，CoDNet的有效性率为99.02%，优于现有的最先进方法。这种具有竞争力的性能突出了CoDNet药物设计的精度和效能，使其成为药物开发计划中具有巨大潜力的重大进展。
