An overview of recent advances and challenges in predicting compound-protein interaction (CPI)

Abstract

Compound-protein interactions (CPIs) are critical in drug discovery for identifying therapeutic targets, drug side effects, and repurposing existing drugs. Machine learning (ML) algorithms have emerged as powerful tools for CPI prediction, offering notable advantages in cost-effectiveness and efficiency. This review provides an overview of recent advances in both structure-based and non-structure-based CPI prediction ML models, highlighting their performance and achievements. It also offers insights into CPI prediction-related datasets and evaluation benchmarks. Lastly, the article presents a comprehensive assessment of the current landscape of CPI prediction, elucidating the challenges faced and outlining emerging trends to advance the field.

Keywords: artificial intelligence; chemogenomics; compound-protein interaction prediction; drug discovery; scoring function.

Figure 1:

Categorization of compound-protein interaction (CPI) prediction models. FASTA, FAST-All; SMILES, Simplified molecular input line entry system.

Figure 2:

Feature engineering-based strategies. (A) Specific energy terms. (B) Protein-ligand atom pairwise counts. (C) Protein-ligand interaction fingerprints. (D) Mathematical features.

Figure 3:

Feature learning-based representation strategies. (A) Atom context-based features. (B) Grid-based features. (C) Graph-based features.

Figure 4:

Non-structure-based CPI prediction methodology. (A) CPI prediction model pipeline; (B) architecture of convolutional neural network; (C) architecture of recurrent neural network; (D) architecture of graph neural network; (E) three common input types of models: proteins are input as sequence and small molecules are constructed as graphs; protein pockets and small molecules are constructed into different graphs; protein pockets and small molecules are converted into the same graphs.