Molecular descriptors have been used to characterize and predict the functions of small molecules, including inhibitors of protein–protein interactions (iPPIs). Such molecules are valuable to investigate disease pathways and as starting points for drug discovery endeavors. iPPIs tend to bind at the surface of macromolecules and the design of such compounds remains challenging. Here, we report on our investigation of a pool of interpretable molecular descriptors for solvent-exposed and buried co-crystallized ligands. Several descriptors were found to be significantly different between the two classes and were further exploited using machine-learning approaches. This work could open new perspectives for the rational design of focused libraries enriched in new types of small drug-like molecules that could be used to prevent PPIs.

分子描述符已用于表征和预测小分子的功能，包括蛋白质间相互作用（iPPIs）的抑制剂。这样的分子对于研究疾病途径和作为药物发现努力的起点是有价值的。iPPI倾向于结合在大分子的表面，此类化合物的设计仍然具有挑战性。在这里，我们报告我们对溶剂暴露和掩埋共结晶配体的可解释分子描述符库的研究。发现两个类之间的几个描述符存在显着差异，并使用机器学习方法进一步加以利用。这项工作可以为聚焦库的合理设计开辟新的前景，该库富含可用于预防PPI的新型小药物样小分子。