Does a More Precise Chemical Description of Protein–Ligand Complexes Lead to More Accurate Prediction of Binding Affinity?

TLDR

Predicting binding affinities for diverse molecules is challenging, and classical scoring functions rely on predetermined functional forms that struggle to model intermolecular contributions, making them essential yet limited tools for docking and drug design. The study investigates how the chemical description of protein–ligand complexes affects the predictive accuracy of scoring functions through systematic numerical experiments. The authors conduct systematic numerical experiments and analyze four factors—modeling assumptions, representation–regression codependence, bound‑state data limitation, and conformational heterogeneity—to explain the observed results. Machine‑learning scoring functions surpass state‑of‑the‑art benchmarks, yielding the most accurate model to date, but a more precise chemical description of protein–ligand complexes does not consistently enhance binding‑affinity prediction accuracy.

Abstract

Predicting the binding affinities of large sets of diverse molecules against a range of macromolecular targets is an extremely challenging task. The scoring functions that attempt such computational prediction are essential for exploiting and analyzing the outputs of docking, which is in turn an important tool in problems such as structure-based drug design. Classical scoring functions assume a predetermined theory-inspired functional form for the relationship between the variables that describe an experimentally determined or modeled structure of a protein–ligand complex and its binding affinity. The inherent problem of this approach is in the difficulty of explicitly modeling the various contributions of intermolecular interactions to binding affinity. New scoring functions based on machine-learning regression models, which are able to exploit effectively much larger amounts of experimental data and circumvent the need for a predetermined functional form, have already been shown to outperform a broad range of state-of-the-art scoring functions in a widely used benchmark. Here, we investigate the impact of the chemical description of the complex on the predictive power of the resulting scoring function using a systematic battery of numerical experiments. The latter resulted in the most accurate scoring function to date on the benchmark. Strikingly, we also found that a more precise chemical description of the protein–ligand complex does not generally lead to a more accurate prediction of binding affinity. We discuss four factors that may contribute to this result: modeling assumptions, codependence of representation and regression, data restricted to the bound state, and conformational heterogeneity in data.