Microtubule associated proteins (MAPs) are crucial for regulating microtubule (MT) dynamics, a process essential for neuronal health. Their structural diversity—ranging from intrinsically disordered proteins (IDPs) such as tau, to MAPs with well-defined folded domains such as severing proteins—leads to distinct regulatory functions. In this work, we focus on three neuronal MAPs: tau, MAP7, and double-cortin (DCX). A fundamental mechanism shared by these three MAPs is their ability to form envelopes on MTs that restrict the binding of severing enzymes to MTs. Because the stability and regulatory role of these envelopes depend directly on the strength of the MAP-MT interactions, it is crucial to calculate the binding energies that govern these interactions. Conventional free energy calculation methods are computationally demanding and poorly suited for the conformational heterogeneity of IDPs, such as tau and MAP7. We developed a computational framework that integrates molecular dynamics (MD) simulations with a machine learning ensemble inspired by the protein binding energy estimator (PBEE). The model utilizes a super learner ensemble that combines predictions from ten distinct machine-learning classifiers. While the PBEE method provides an efficient approach to estimating binding energies, it has several shortcomings: the original data set was relatively small with a negligible number of IDPs, and the method was designed for single (PDB) protein structures. To address these issues, we expanded the original data set to ∼3,200 complexes, providing better coverage of both IDPs and folded proteins, and extracted the binding energy from MD trajectories. Using tau, MAP7, and DCX as representative MAPs, we sampled conformations from MD trajectories and extracted Rosetta-based features to determine the respective binding free energies. Our methodology leads to results consistent with experimental observations that support the specific function of each MAP.
Irshad et al. (Sun,) studied this question.