An Improvement of the Free Energy Perturbation (FEP+) Sampling Protocol in Case of Flexible protein Ligand Binding Sites
Micar21 Ltd., Persenk Str. 34B, 1407 Sofia, Bulgaria
Received: ... 2018; Accepted: ... 2018; Accepted author version posted online: ... 2018
The recent improvements of the free energy perturbation calculations (FEP) and in particular the FEP+ protocol established the FEP approach as a useful tool during the hit to lead optimization process. However, this method is manly helpful when a good quality X-ray data is available and the target protein dose not undergo significant conformational changes. The lack of systematic studies on adjustment of an adequate equilibration sampling time is one of the main issues during the FEP calculations in many cases. In a current study we propose modified version of the FEP+ sapling protocol which was created by probing large number of permutations with different sampling schemes. Essentially, we found that extending the pre-equilibration time from 0.24ns to 5ns and 2x10ns can greatly improve the FEP+ calculations in a case of regular flexible loops monitions and significant structural changes, respectively. The FEP/REST simulations were also extended from 5ns to 8ns in order a reasonable free energy convergence to be achieved. The implementation in the REST region of the whole ligand, instead of only perturbation portion, and also some basic for the protein flexibility residues (pREST region) improved further significantly the FEP+ results in most of the studied cases. Much more precise calculations of the ddG values of the individual perturbations were also achieved. The preliminary Molecular dynamics (MD) runs were also of great help to establish the correct binding mode of compounds and thus to align them precisely for a FEP+ execution. Our protocol was initially developed and tested on group of PPARγ partial agonists and MD derived complexes demonstrating that the FEP+ can be successfully used alongside with such structures, which is very important in a case of homology modeling improvements. Further, we validated our workflow on different targets including three which have been already intensively studied. In particular, we demonstrated that in a case of T4 Lysozyme L99A the insufficient sampling caused due to the significant structural rearrangements, which were previously overcome by 55ns FEP/pREST simulations, can be fixed by 2 times less execution time using our protocol keeping the error value the same as in pREST procedure (RMSE=0.54 kcal/mol). Notably, this indicates that the pre-equilibration is a critical step before execution of the actual FEP calculations. Impressive results and an improvement compared to the previous calculations were also obtained for both the Thrombin and Tyk2 systems. The RMSE value for the test set of Tyk2 ligands was decreased from 0.93 kcal/mol using the standard protocol to 0.53 kcal/mol by our sampling workflow. The much higher prediction levels were also evident from calculated R2 values. For instance, in the case of Thrombin set of compounds the R2 value increased from 0.58 to 0.87. Finally, we concentrated on series of AKT1 inhibitors, which substitutions were in a different positions and improved the RMSE more than 3 times compared to the standard FEP+ sampling protocol (MUE=0.54, RMSE=0.66 and R2=0.73) making the use of the FEP+ approach independent of the starting structure in that case too. Based on these data we hope that our workflow can be commonly adopted because it significantly improved the FEP+ results in all studied cases.