Structure-based drug repositioning has emerged as a promising alternative to conventional drug development. Regardless of the many success stories reported over the past years and the novel breakthroughs on the AI-based system AlphaFold for structure prediction, the availability of structural data for protein–drug complexes remains very limited. Whereas the chemical libraries contain millions of drug compounds, the vast majority of them do not have structures to crystallized targets,and it is, therefore, impossible to characterize their binding to targets from a structural view. However, the concept of building blocks offers a novel perspective on the structural problem. A drug compound is considered a complex of small chemical blocks or fragments, which confer the relevant properties to the drug and have a high proportion of functional groups involved in protein binding. Based on this, we propose a novel approach to expand the scope of structure-based repositioning approaches by transferring the structural knowledge from a fragment to a compound level. We fragmented over 100,000 compounds in the Protein Data Bank (PDB) and characterized the structural binding mode of 153,000 fragments to their crystallized targets. Using the fragment’s data, we were able to artificially reconstruct the binding mode of over 7,800 complexes between ChEMBL compounds and their known targets, for which no structural data is available. We proved that the conserved binding tendency of fragments, when binding to the same targets, highly influences the drug’s binding specificity and carries the key information to reconstruct full drugs binding mode. Furthermore, our approach was able to reconstruct multiple compound-target pairs at optimal thresholds and high similarity to the actual binding mode. Such reconstructions are of great value and benefit structure-based drug repositioning since they automatically enlarge the technique’s scope and allow exploring the so far ‘unexplored compounds’ from a structural perspective. In general, the transfer of structural information is a promising technique that could be applied to any chemical library, to any compound that has no crystal structure available in PDB, and even to transfer any other feature that may be relevant for the drug discovery process and that due to data limitations is not yet fully available. In that sense, the results of this work document the full potential of structure-based screening even beyond PDB.

中文翻译：

---

分解化合物能够重建相互作用指纹，用于基于结构的药物筛选

基于结构的药物重新定位已成为传统药物开发的有希望的替代方案。尽管过去几年报道了许多成功案例，以及基于人工智能的系统 AlphaFold 结构预测的新突破，但蛋白质-药物复合物的结构数据的可用性仍然非常有限。尽管化学文库包含数百万种药物化合物，但其中绝大多数不具有与结晶靶标的结构，因此不可能从结构角度表征它们与靶标的结合。然而，积木的概念为结构问题提供了一个新的视角。药物化合物被认为是小化学块或片段的复合物，它们赋予药物相关特性，并具有高比例的参与蛋白质结合的官能团。基于此，我们提出了一种新方法，通过将结构知识从片段转移到复合水平来扩展基于结构的重新定位方法的范围。我们对蛋白质数据库 (PDB) 中的 100,000 多种化合物进行了片段化，并对 153,000 个片段与其结晶靶标的结构结合模式进行了表征。使用该片段的数据，我们能够人工重建 7,800 多种 ChEMBL 化合物与其已知靶标之间的结合模式，但没有可用的结构数据。我们证明了片段的保守结合趋势，当与相同的目标结合时，高度影响药物的结合特异性，承载着重构药物全结合模式的关键信息。此外，我们的方法能够以最佳阈值和与实际结合模式的高度相似性重建多个化合物-目标对。这种重建具有巨大的价值和有益于基于结构的药物重新定位，因为它们自动扩大了技术的范围，并允许从结构的角度探索迄今为止“未开发的化合物”。一般来说，结构信息的转移是一种很有前途的技术，可以应用于任何化学库、任何在 PDB 中没有可用晶体结构的化合物，甚至可以转移任何其他可能与药物发现过程相关的特征和由于数据限制，尚未完全可用。从这个意义上说，