Knowledge Based Potentials: the Reverse Boltzmann Methodology, Virtual Screening and Molecular Weight Dependence

Abstract

We discuss the rationale for using the reverse Boltzmann methodology to convert atomatom distance distributions from a knowledge base of protein-ligand complexes into energy-like functions. We also generate an updated version of the BLEEP statistical potential, using a dataset of 196 complexes. This performs similarly to the existing BLEEP. An algorithm is implemented to allow the automatic calculation of bond orders, and hence of the appropriate numbers of hydrogen atoms present. An attempt is made to generate a potential specific to strongly bound complexes; however, we find no evidence that this improves the prediction of binding affinities. We also discuss the range of binding energies available as a function of ligand molecular weight and derive some simple functions describing this behaviour.

Druggability Assessment of Allosteric Proteins by Dynamics Simulations in the Presence of Probe Molecules

“…It should be noted that despite its broad use in docking and folding studies, the physical basis of inverse Boltzmann relation for relative free energy calculations is debated. 31 Concerns usually arise in cases where interacting particles are not independent, such as covalently linked amino acids in a protein. 32 However, situations where it is clearly applicable for relative free energy calculations include equilibrium systems of independent particles.…”

“… 32 However, situations where it is clearly applicable for relative free energy calculations include equilibrium systems of independent particles. 31 Thus, we accounted for binding events by considering probe molecules as whole independent particles. The resulting relative binding free energy map (Figure 1 D) is refined to identify interaction spots (Figure 1 E), each representing a probe molecule.…”

Ensemble docking to difficult targets in early‐stage drug discovery: Methodology and application to fibroblast growth factor 23

“…There is a general trend for compounds of higher molecular weight to have higher scores in the present screens. The tendency of virtual screening procedures to be biased toward higher molecular weights is well known . Here, however, we report the raw scores without any renormalization, contrary to the prescriptions of others .…”

A Random Forest Model for Predicting Allosteric and Functional Sites on Proteins

“…Previous knowledge‐based approaches used ensembles of observed protein‐ligand crystal structures to infer binding energies from atom‐atom distance distributions. That approach makes the dubious assertion that Boltzmann energetics apply, assuming a particular exponential functional form to transform distance distributions into binding energies 27. RF‐Score uses Random Forest to predict binding affinities from both structural data and the affinity data that are left unused in most knowledge‐based approaches, yielding a much more accurate and flexible scoring function.…”