Search

We have recently developed a 2P2IHUNTER learning machine tool for filtering potential orthosteric PPI modulators. In the present research article we applied our algorithm to 8.3 million compounds representing the main chemical providers commercially available to design a PPI-focused library, 2P2IREF, composed of 143 218 small molecules. Compounds corresponding to medicinally important privileged structures identified as core structures in numerous therapeutics were prioritized in a medicinal oriented version of this database (2P2IPRIV – 51 476 compounds). A diverse chemical library was generated using 2D molecular fingerprints (2P2IDIV – 8217 diverse compounds). The carbon bond saturation index (Fsp3) was used as a final filter (Fsp3 ≥ 0.4) to escape from flatland, another hurdle in the long path to the gold mine. We analyzed the resulting chemical space of this final library of 1683 compounds, 2P2I3D and discuss the chemical moieties proposed to the community. This library will now be tested to evaluate its ability to enhance hit rates in general screening campaigns and is open to academic and private companies for collaborative prospects.

Protein-protein interactions (PPIs) represent an enormous source of opportunity for therapeutic intervention. We and others have recently pinpointed key rules that will help in identifying the next generation of innovative drugs to tackle this challenging class of targets within the next decade. We used these rules to design an oriented chemical library corresponding to a set of diverse 'PPI-like' modulators with cores identified as privileged structures in therapeutics. In this work, we purchased the resulting 1664 structurally diverse compounds and evaluated them on a series of representative protein-protein interfaces with distinct "druggability" potential using Homogeneous Time-Resolved Fluorescence (HTRF®) technology. For certain PPI classes, analysis of the hit rates revealed up to 100 enrichment factors compared with non-oriented chemical libraries. This observation correlates with the predicted "druggability" of the targets. A specific focus on selectivity profiles, the three-dimensional (3D) molecular modes of action resolved by X-ray crystallography, and the biological activities of identified hits targeting the well-defined "druggable" bromodomains of the bromo and extraterminal (BET) family are presented as a proof-of-concept. Overall, our present study illustrates the potency of machine learning-based oriented chemical libraries to accelerate the identification of hits targeting PPIs. A generalization of this method to a larger set of compounds will accelerate the discovery of original and potent probes for this challenging class of targets.

Over the last 10 years, protein–protein interactions (PPIs) have shown increasing potential as new therapeutic targets. As a consequence, PPIs are today the most screened target class in high-throughput screening (HTS). The development of broad chemical libraries dedicated to these particular targets is essential; however, the chemical space associated with this ‘high-hanging fruit’ is still under debate. Here, we analyse the properties of 40 non-redundant small molecules present in the 2P2I database ( http://2p2idb.cnrs-mrs.fr/ ) to define a general profile of orthosteric inhibitors and propose an original protocol to filter general screening libraries using a support vector machine (SVM) with 11 standard D ragon molecular descriptors. The filtering protocol has been validated using external datasets from PubChem BioAssay and results from in-house screening campaigns . This external blind validation demonstrated the ability of the SVM model to reduce the size of the filtered chemical library by eliminating up to 96% of the compounds as well as enhancing the proportion of active compounds by up to a factor of 8. We believe that the resulting chemical space identified in this paper will provide the scientific community with a concrete support to search for PPI inhibitors during HTS campaigns.