TAGMol: Target-Aware Gradient-guided Molecule Generation

3D生成模型在基于结构的药物设计中表现出了显著的潜力，特别是在发现适合特定靶点结合位点的配体方面。现有的算法通常主要关注配体-靶点的结合亲和力。此外，仅基于靶点-配体分布训练的模型可能无法满足药物发现的更广泛目标，如开发具有所需性质（如药物样性和可合成性）的新型配体，突显了药物设计过程的多面性。为了克服这些挑战，我们将问题分解为分子生成和性质预测两个部分。后者协同地指导扩散采样过程，促进有导向的扩散，从而创造具有所需性质的有意义的分子。我们将这个有导向的分子生成过程称为TAGMol。通过对基准数据集的实验，TAGMol表现出比最先进的基线模型更优异的性能，平均Vina得分提高了22％，并在必要的辅助属性方面产生了有利的结果。这将TAGMol确立为药物生成的综合框架。

TAGMol: A Comprehensive Framework for Structure-Based Drug Design Using Generative Models

TAGMol decouples the problem of structure-based drug design into molecular generation and property prediction, using the latter to guide the diffusion sampling process and create meaningful molecules with desired properties.

TAGMol achieves a 22% improvement in average Vina Score compared to state-of-the-art baselines and yields favorable outcomes in essential auxiliary properties. Experiments were conducted on benchmark datasets, and the framework is comprehensive and open-source.

Related work includes the use of generative models in structure-based drug design, such as GENTRL and MolDQN, as well as research on property prediction and optimization in drug design, such as the work on QED and synthetic accessibility.