Quantitative Structure-Activity Relationships (QSAR) typically rely on the two- and three-dimensional structures of molecules to predict their bioactivity. These models have proven reliable for virtual screening of large chemical libraries but have also shown poor performances when it comes to lead optimization and assisting medicinal chemists. Herein, we introduce the MD-QSAR modeling approach that uses machine learning and descriptors computed from the molecular dynamics trajectories of kinase-inhibitor complexes. We will discuss the rationale of the approach, its origin with early attempts of 4D-QSAR modeling, and two case studies involving a set of 85 ERK2 kinase inhibitors and another set of 925 Bcr-Abl tyrosine-kinase inhibitors (imatinib analogues). The MD-QSAR modeling workflow includes (i) the structure-based docking of all compounds in the binding site of the kinase, (ii) the independent molecular dynamics (MD) simulations of each protein-ligand complex (Desmond-GPU, 15 ns, NTP, 300K, TIP3P, 1fs), (iii) the computation of MD fingerprints to characterize ligands’ conformational flexibility and the dynamic kinase-inhibitor interactions over the trajectories, and (iv) the training and cross-validation of MD-QSAR models using machine learning techniques (random forests and artificial neural networks) and the aforementioned MD descriptors. MD-QSAR models afforded better prediction performances compared to classical 2D and 3D QSAR models. Moreover, the interpretation of MD descriptors is facilitated with the direct visualization of the dynamic kinase-inhibitor interactions. We posit this new modeling workflow combining machine learning, 3D docking, and molecular dynamics simulations can provide key knowledge for the design of more potent and selective analogues.

MD-QSAR modeling of dynamic kinase-inhibitor interactions using machine learning and molecular dynamics trajectories

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.