Evaluating the use of absolute binding free energy in the fragment optimisation process

Abstract

Key to the fragment optimisation process within drug design is the need to accurately capture the changes in affinity that are associated with a given set of chemical modifications. Due to the weakly binding nature of fragments, this has proven to be a challenging task, despite recent advancements in leveraging experimental and computational methods. In this work, we evaluate the use of Absolute Binding Free Energy (ABFE) calculations in guiding fragment optimisation decisions, retrospectively calculating binding free energies for 59 ligands across 4 fragment elaboration campaigns. We first demonstrate that ABFEs can be used to accurately rank fragment-sized binders with an overall Spearman's r of 0.89 and a Kendall τ of 0.67, although often deviating from experiment in absolute free energy values with an RMSE of 2.75 kcal/mol. We then also show that in several cases, retrospective fragment optimisation decisions can be supported by the ABFE calculations. Comparing against cheaper endpoint methods, namely N_wat-MM/GBSA, we find that ABFEs offer better ranking power and correlation metrics. Our results indicate that ABFE calculations can usefully guide fragment elaborations to maximise affinity.

Fig. 1. Overview of the fragment elaboration datasets.

A total of 59 ligands from four different elaboration studies^– with affinities spanning the millimolar to nanomolar range are investigated here.

Fig. 2. Absolute binding free energy thermodynamic cycle employed.

The free energy of binding, i.e. going from a ligand in solution (state a) to a protein-ligand complex (state e), is captured through a non-physical path. First, the electrostatics are annihilated to zero (state b) over 11 λ windows. This is followed by a further 21 λ windows which decouple the ligand van der Waals interaction from the solvent (state c). The decoupled ligand is then analytically restrained as defined by ref. (state d). By accounting for this restraint the ligand state is then equivalent to a non-interacting ligand in a protein-ligand complex (state h). The ligand interactions with the environment are then turned back on, first re-coupling the van der Waals interactions over 21 λ windows (state g), followed by a further 11 λ windows to add back electrostatics (state f). Finally, the orientational restraints are turned off over 12 λ windows resulting in a fully interacting protein-ligand complex (state e).

Fig. 3. ABFE calculations results for different systems.

a All four datasets, b PWWP1, c HSP90, d MCL-1, e Cyclophilin D. Free energy estimates are the means of the estimates across replicas, with error bars as their standard deviation. Correlation metrics calculated from the mean estimate values, with error bars derived from bootstrap resampling. All free energy results, including RMSE values, have units of kcal/mol.

Fig. 4. Impact of conformation on ABFE calculations for HSP90.

a Comparison of the structures of PDB IDs 2XHT (blue) and 2XDL (red) demonstrating the difference in helicity between the two models and b, c a zoomed view of the affected helix. Residues 100–124 have been highlighted in a darker colour to aid in visualisation. PDB ID 2XAB, which is similar in structure to 2XDL and 2XHX which is similar in structure to 2XHT are not shown. d Absolute binding free energies of the HSP90 ligands all starting from the 2XDL-like helix loop conformation. Free energy estimates are the means of the estimates across replicas, with error bars as their standard deviation. Correlation metrics calculated from the mean estimate values, with error bars derived from bootstrap resampling. All free energy results, including RMSE values, have units of kcal/mol.

Fig. 5. Overlay of the starting configurations of each replica of the ABFE calculations for ligand 60 in MCL-1.

The re-arrangement (pink coloured ligand) of the merged ligand 60 in the MCL-1 binding site can clearly be seen.

Fig. 6. Binding free energies estimated by N_wat-MM/GBSA.

a PWWP1, b HSP90, c MCL-1, and d Cyclophilin D datasets. Free energy estimates are the means of the estimates across replicas, with error bars as their standard deviation. Correlation metrics calculated from the mean estimate values, with error bars derived from bootstrap resampling. All free energy results, have units of kcal/mol.

Fig. 7. Comparison of a subset of the MCL-1 dataset.

a ABFE simulations and b the 2015 FEP + study by ref. Free energy estimates are the means of the estimates across replicas, with error bars as their standard deviation. Correlation metrics calculated from the mean estimate values, with error bars derived from bootstrap resampling. All free energy results, including RMSE values, have units of kcal/mol.