ML‑Driven de novo Design and in Silico Validation of ERK2 Leads (Ek1–Ek4)

An MDPI article describes a machine learning–assisted de novo design and in silico validation workflow that produced four candidate ERK2 modulators labeled Ek1–Ek4.

The authors present a staged computational triage that moves from initial molecule generation through successive binding and developability screens, including molecular dynamics simulation, with free-energy perturbation (FEP) reported as the final binding-affinity evaluation step. The sequence is described as progressing from docking to pose refinement, then to dynamic simulation, free-energy perturbation, and an in silico pharmacokinetics/medicinal-chemistry review.

The study outlines a stepwise filtering pipeline that begins with molecule generation using DeLA‑Drug, followed by a similarity search against active ERK2 inhibitors to narrow chemical space. Remaining candidates were evaluated with molecular docking using AutoDock Vina, and the authors report an additional pharmacokinetic assessment performed with DeepPK. For the final selected set, the workflow describes pose refinement using DiffDock before molecular dynamics simulations for dynamical assessment. The authors also report performing free-energy perturbation calculations in Gromacs2023.4 as a later-stage binding-affinity exploration. Overall, the pipeline is presented as a multi-tier computational prioritization process that narrows from a larger generated pool to four leads.

For docking, the authors report that the selected molecules showed higher predicted binding affinity for ERK2, with docking scores ranging from −9.50 to −10.50 kcal/mol using AutoDock Vina. DiffDock is reported as the tool used to refine poses for the final selected molecules prior to downstream simulation-based analyses. These docking and pose-handling steps are framed as an initial screen supporting selection of Ek1–Ek4 for more computationally intensive evaluation. With top-ranked docking results in hand, the study then turns to dynamic behavior and free-energy calculations as additional checks on association with ERK2.

In describing molecular dynamics, the authors report that dynamics assessment supported a strong association of the screened molecules with ERK2, attributing this to lower deviation of the ERK2 backbone in dynamic states. The report positions these MD observations as corroborative evidence alongside earlier docking-based selection, without detailing additional simulation settings in the abstract text. For free-energy perturbation, the authors report an Ek1 FEP energy of −26.85 kJ/mol compared with −22.77 kJ/mol for a standard molecule, which they interpret as indicating strong affinity of Ek1 toward ERK2. In the study narrative, MD and FEP are presented as binding-strength checks applied after pose refinement.

Beyond binding-focused computations, the article reports in silico developability-oriented screening, including pharmacokinetic assessment using DeepPK and a review of medicinal-chemistry attributes. The authors state that all four screened molecules had satisfactory pharmacokinetic properties, satisfactory medicinal-chemistry properties, and good synthetic accessibility scores, describing these features as consistent with drug-like compounds under Lipinski’s rule of five.

In their conclusion, the authors characterize the proposed ERK2 modulators as potential avenues for future drug discovery targeting ERK2, while emphasizing that experimental validation remains necessary. Overall, the study presents Ek1–Ek4 as computationally prioritized ERK2-focused leads generated and assessed through a defined multi-step in silico workflow.

Key Takeaways:

• The authors describe a DeLA‑Drug-to-validation sequence that includes similarity filtering, docking, pharmacokinetic assessment, pose refinement, MD simulation, and FEP calculations.
• Across docking, MD, and FEP stages, the study reports higher docking binding affinity for selected molecules, describes a strong association with ERK2 in MD simulation, and reports an Ek1 FEP value that the authors interpret as indicating strong affinity toward ERK2.
• The authors conclude that the candidates represent potential avenues for ERK2-focused discovery, while noting that experimental validation is required.