Improved Accuracy for Modeling PROTAC-Mediated Ternary Complex Formation and Targeted Protein Degradation via New In Silico Methodologies.

Extending upon our previous publication [Drummond, M.; J. Chem. Inf. Model. 2019, 59, 1634-1644], two additional computational methods are presented to model PROTAC-mediated ternary complex structures, which are then used to predict the efficacy of any accompanying protein degradation. Method 4B, an extension to one of our previous approaches, incorporates a clustering procedure uniquely suited for considering ternary complexes. Method 4B yields the highest proportion to date of crystal-like poses in modeled ternary complex ensembles, nearing 100% in two cases and always giving a hit rate of at least 10%. Techniques to further improve this performance for particularly troublesome cases are suggested and validated. This demonstrated ability to reliably reproduce known crystallographic ternary complex structures is further established through modeling of a newly released crystal structure. Moreover, for the far more common scenario where the structure of the ternary complex intermediate is unknown, the methods detailed in this work nonetheless consistently yield results that reliably follow experimental protein degradation trends, as established through seven retrospective case studies. These various case studies cover challenging yet common modeling situations, such as when the precise orientation of the PROTAC binding moiety in one (or both) of the protein pockets has not been experimentally established. Successful results are presented for one PROTAC targeting many proteins, for different PROTACs targeting the same protein, and even for degradation effected by an E3 ligase that has not been structurally characterized in a ternary complex. Overall, the computational modeling approaches detailed in this work should greatly facilitate PROTAC screening and design efforts, so that the many advantages of a PROTAC-based degradation approach can be effectively utilized both rapidly and at reduced cost.

In Silico Modeling of PROTAC-Mediated Ternary Complexes: Validation and Application.

In this work, four methods are described and validated for generating in silico ensembles of PROTAC-mediated ternary complexes. Filters based on characteristics of the proposed ternary complexes are developed to identify those that resemble known crystal structures. We then show how to use these modeling techniques a priori to discriminate the PROTAC-mediated degradation behavior of a mutant protein vs its wild type, of three closely related targets, and among three different PROTAC molecules.

Numerical errors in minimization based binding energy calculations.

This work examines the effect of small input perturbations on binding energies computed from differences between energy minimized structures, such as the Prime MM-GBSA and MOE MM-GB/VI methods. The applied perturbations include translations of the cognate ligand in the binding site by a maximum of 0.1 Å along each coordinate or the permutation of the order of atoms of the cognate ligand without any changes to the atom coordinates. These seemingly inconsequential input changes can lead to as much as 17 kcal/mol differences in the computed binding energy. The calculated binding energies cluster around discrete values, which correspond to specific ligand poses. It appears that the largest variations are observed for target-ligand systems in which there is a possibility for multiple poses with strong hydrogen bonds. The barriers between different poses can appear fractal-like, making it difficult to predict which solution will be produced from a given input. Including protein flexibility in MM-GBSA calculations further increases numerical instability, and the protein strain terms seem to be the major factor contributing to this sensitivity. In such calculations it appears unwise to extend the flexible region beyond 6 Å.

Variability in docking success rates due to dataset preparation.

The results of cognate docking with the prepared Astex dataset provided by the organizers of the "Docking and Scoring: A Review of Docking Programs" session at the 241st ACS national meeting are presented. The MOE software with the newly developed GBVI/WSA dG scoring function is used throughout the study. For 80 % of the Astex targets, the MOE docker produces a top-scoring pose within 2 Å of the X-ray structure. For 91 % of the targets a pose within 2 Å of the X-ray structure is produced in the top 30 poses. Docking failures, defined as cases where the top scoring pose is greater than 2 Å from the experimental structure, are shown to be largely due to the absence of bound waters in the source dataset, highlighting the need to include these and other crucial information in future standardized sets. Docking success is shown to depend heavily on data preparation. A "dataset preparation" error of 0.5 kcal/mol is shown to cause fluctuations of over 20 % in docking success rates.

Numerical errors and chaotic behavior in docking simulations.

This work examines the sensitivity of docking programs to tiny changes in ligand input files. The results show that nearly identical ligand input structures can produce dramatically different top-scoring docked poses. Even changing the atom order in a ligand input file can produce significantly different poses and scores. In well-behaved cases the docking variations are small and follow a normal distribution around a central pose and score, but in many cases the variations are large and reflect wildly different top scores and binding modes. The docking variations are characterized by statistical methods, and the sensitivity of high-throughput and more precise docking methods are compared. The results demonstrate that part of docking variation is due to numerical sensitivity and potentially chaotic effects in current docking algorithms and not solely due to incomplete ligand conformation and pose searching. These results have major implications for the way docking is currently used for pose prediction, ranking, and virtual screening.

Reducing docking score variations arising from input differences.

The variability of docking results as a function of variations in ligand input conformations was studied for the GOLD, Glide, FlexX, and Surflex programs. It is concluded that there are two major effects leading to such variability: the adequacy of conformational search during docking and random "chaotic" effects arising from sensitivity to small input perturbations. It is shown that although the former is generally the stronger effect, the latter is also highly significant for almost all docking engines. The strong target-to-target variation of the magnitude of these effects is emphasized. The performance of different packages is compared using these measures. Guidelines are provided for different programs to reduce variability and improve reproducibility, which involve using a small number of input conformations as starting points for docking, followed by the selection of the top scoring docked pose from the results as the best docked solution.

Effect of input differences on the results of docking calculations.

The sensitivity of docking calculations to the geometry of the input ligand was studied. It was found that even small changes in the ligand input conformation can lead to large differences in the geometries and scores of the resulting docked poses. The accuracy of docked poses produced from different ligand input structures-the X-ray structure, the minimized Corina structure, and structures generated from conformational searches and molecular dynamics ensembles-were also assessed. It was found that using the X-ray ligand conformation as docking input does not always produce the most accurate docked pose when compared with other sources of ligand input conformations. Furthermore, no one method of conformer generation is guaranteed to always produce the most accurate docking pose. The docking scores are also highly sensitive to the source of the input conformation, which might introduce some noise in compound ranking and in binding affinity predictions. It is concluded that for the purposes of reproducibility and optimal performance, the most prudent procedure is to use multiple input structures for docking. The implications of these results on docking validation studies are discussed.

The effect of numerical error on the reproducibility of molecular geometry optimizations.

Geometry optimization is one of the most often applied techniques in computational drug discovery. Although geometry optimization routines are generally deterministic, the minimization trajectories can be extremely sensitive to initial conditions, especially in case of larger systems such as proteins. Simple manipulations such as coordinate transformations (translations and rotations), file saving and retrieving, and hydrogen addition can introduce small variations ( approximately 0.001 A) in the starting coordinates which can drastically affect the minimization trajectory. With large systems, optimized geometry differences of up to 1 A RMSD and final energy differences of several kcal/mol can be observed when using many commercially available software packages. Differences in computer platforms can also lead to differences in minimization trajectories. Here we demonstrate how routine structure manipulations can introduce small variations in atomic coordinates, which upon geometry optimization, can give rise to unexpectedly large differences in optimized geometries and final energies. We also show how the same minimizations run on different computer platforms can also lead to different results. The implications of these findings on routine computational chemistry procedures are discussed.