pdCSM-PPI: Using Graph-Based Signatures to Identify Protein–Protein Interaction Inhibitors

Abstract

Protein−protein interactions are promising sites for development of selective drugs; however, they have generally been viewed as challenging targets. Molecules targeting protein−protein interactions tend to be larger and more lipophilic than other druglike molecules, mimicking the properties of interacting interfaces. Here, we propose a machine learning approach that uses a graphbased representation of small molecules to guide identification of inhibitors modulating protein−protein interactions, pdCSM-PPI. This approach was applied to 21 different PPI targets. We developed interaction-specific models that were able to accurately identify active compounds achieving MCC and F1 scores up to 1, and Pearson's correlations up to 0.87, outperforming previous approaches. Using insights from these individual models, we developed a generic protein−protein interaction modulator predictive model, which accurately predicted IC50 with a Pearson's correlation of 0.64 on a low redundancy blind test. Importantly, we were able to accurately identify active from inactive compounds, achieving an AUC of 0.77 and sensitivity and specificity of 76% and 78%, respectively. We believe pdCSM-PPI will be an important tool to help guide more efficient screening of new PPI inhibitors; it is freely available as an easy-to-use web server and API at http:// biosig.unimelb.edu.au/pdcsm_ppi.

cardioToxCSM: A Web Server for Predicting Cardiotoxicity of Small Molecules

“…Multiple predictive models were developed using supervised learning, with classification as the prediction task (e.g., all data sets had categorical outcomes: cardiotoxic/cardiosafe). During feature engineering, different classes of properties were investigated including graph-based signatures, , molecular fingerprints, and 2D descriptors . The best performing model involved a combination of graph-based signatures and molecular fingerprints and was thus selected for further evaluation.…”

“…Graph-based signatures are a well-established modeling tool to represent chemical compounds and their physicochemical and geometrical properties in order to understand their structure–activity relationships . These have been applied successfully in a range of scenarios, including modeling for effects of mutations on proteins and their partners and macromolecule properties and to model bioactivity, pharmacokinetics, and toxicity. , Here, we have utilized these signatures to model small-molecule properties for the prediction of cardiotoxic compounds. In each of these graphs, nodes represent atoms (labels based on their pharmacophoric properties), while edges between them show covalent bonding.…”

A Multimodal Deep Learning Framework for Predicting PPI-Modulator Interactions

“…SMMPPI adopts a two-stage classification method to predict PPIMs based on molecular structure fingerprints . pdCSM-PPI utilizes a graph-based compound representation to facilitate the discovery of PPIMs for 21 different PPI targets . However, existing ML- or DL-based methods predict modulators based on similarity to known ones, while ignoring PPI target information and the interaction principles between PPI targets and modulators.…”

Predicting the dynamic interaction of intrinsically disordered proteins

“…Protein-protein interactions (PPIs) play a central role in all biological processes [1][2][3] . Conventional PPIs are encoded in the complementarity of proteins' binding interfaces, with the premise of well-defined protein conformation [4][5][6][7][8] . On the other hand, there are numerous proteins that lack stable structures and are denoted as intrinsically disordered proteins (IDPs).…”