All seven human mitogen-activated protein kinase kinases (MAP2Ks) are essential for cellular processes such as cell proliferation and apoptosis, and dysfunction of MAP2Ks is associated with cancers and autoimmune diseases. 5Z-7-oxozeaenol (5Z7O) strongly inhibits MAP2K1, 2, 3, and 6, and weakly inhibits MAP2K4, 5, and 7. In this study, the potencies of 5Z7O toward MAP2K2, 3, and 5 were explored by computational methods including docking and molecular dynamics (MD) simulations using homology models. Docking simulations showed that the α, β-unsaturated ketone moiety of 5Z7O bound close to the conserved cysteine residue located in front of the DFG motif (DFG-1) in MAP2K2 and 3 as in previous studies of MAP2K1 and 6; in strong inhibition, 5Z7O binds covalently with this cysteine. MD simulations of MAP2K2 and 3 in the apo state showed that the high flexibility of a conserved “gatekeeper” methionine residue enabled access of 5Z7O to this binding site. However, in docking simulation of MAP2K5, the α, β-unsaturated ketone moiety of 5Z7O was far from the conserved DFG-1 cysteine. A threonine residue at the gatekeeper position in MAP2K5 likely prevents access of the ketone moiety to that cysteine. These findings, together with previous data, provide guidance for the development of inhibitors selective for each type of MAP2K.

Accurate prediction of protein–ligand binding affinities is a critical step in accelerating drug discovery by reducing experimental costs and development times. Recently developed co-folding AI models predict how multiple biomolecules fold and interact with each other in three-dimensional space. The emergence of a new co-folding model, Boltz-2, has made highly accurate and efficient predictions of protein–ligand binding affinities increasingly feasible. However, the generalization and reliability of these models remain unclear due to the absence of standardized and target-wide benchmark datasets. In this study, we constructed an independent external benchmark dataset derived from ChEMBL version 35 to rigorously evaluate Boltz-2’s performance for affinity prediction. The dataset includes 356 unique protein targets and 10,933 compounds, carefully collected to ensure no overlap with the Boltz-2 training data. Binding affinity measurements were standardized into pChEMBL values and linked to the compound SMILES and protein UniProt accessions. Using this benchmark dataset, we compared the performance of the original Boltz-2 model and its NVIDIA Inference Microservice (NIM) implementation. The results showed that the original Boltz-2 and NIM achieved comparable and fair predictive performance across targets (mean absolute error of approximately 0.9), while NIM reduced computational time by approximately 60–90%. The error analysis indicated that no clear correlation existed between the prediction errors and sequence or compound novelty relative to the Boltz-2 training data, underscoring the model’s broad coverage. This work provides a transparent and reproducible benchmark for evaluating AI-driven affinity prediction models and offers valuable insights into Boltz-2’s applicability, limitations, and potential as a practical tool for data-driven drug discovery.