Advances in Computational Polypharmacology

Abstract

In drug discovery, polypharmacology encompasses the use of small molecules with defined multi-target activity and in vivo effects resulting from multi-target engagement. Multi-target compounds are often efficacious in the treatment of complex diseases involving target and pathway networks, but might also elicit unwanted side effects. Computational approaches such as target prediction or multi-target ligand design have been used to support polypharmacological drug discovery. In addition to efforts directed at the identification or design of new multi-target compounds, other computational investigations have aimed to differentiate such compounds from potential false-positives or explore the molecular basis of multi-target activities. Herein, a concise overview of the field is provided and recent advances in computational polypharmacology through machine learning are discussed.

Keywords: Polypharmacology; computational methods; explainable machine learning; medicinal chemistry; molecular design; multi-target compounds.

Figure 1

X‐ray structures of a multi‐family ligand. The thyronine analogue shown in the center is found in co‐crystal structures with eight targets from five different families, hence providing a prime example for an experimentally confirmed MT‐CPD. On the left, the structure of a complex with human serum albumin is shown (PDB ID: 1HK4) and on the right, a complex with the human thyroid hormone receptor alpha (PDB ID: 4LNX) is illustrated. In these structures, the compound adopted different binding modes.

Figure 2

From feature weights to substructures. Shown is a correctly predicted MT‐CPD. In a), exemplary structural features supporting the prediction (positive SHAP values) and opposing the prediction (negative values) are delineated. While only a few negative contributions were detected, many (partly redundant) fingerprint features made comparably small positive contributions, leading to a high cumulative probability of multi‐target activity (94 %). In b), SHAP values of features present in the compound were mapped to corresponding atoms, highlighting a coherent substructure largely responsible for the correct prediction.

Figure 3

Explainable machine learning in polypharmacology. The compound at the top left shows an exemplary inhibitor with multi‐kinase activity that was correctly predicted via ML. Structural features determining the correct prediction were identified using SHAP analysis and mapped to the inhibitor (second representation following the arrow). These features delineated a coherent substructure (third representation) that distinguished multi‐kinase from single‐kinase inhibitors. This substructure was found in an X‐ray structure of another inhibitor in complex with a kinase (the binding site is enlarged on the right). Following this approach, explainable machine learning has systematically differentiated between inhibitors with multi‐ and single‐kinase activity, leading to experimentally testable hypotheses concerning distinguishing structural motifs.