Mechanism-based organization of neural networks to emulate systems biology and pharmacology models.

Deep learning neural networks are often described as black boxes, as it is difficult to trace model outputs back to model inputs due to a lack of clarity over the internal mechanisms. This is even true for those neural networks designed to emulate mechanistic models, which simply learn a mapping between the inputs and outputs of mechanistic models, ignoring the underlying processes. Using a mechanistic model studying the pharmacological interaction between opioids and naloxone as a proof-of-concept example, we demonstrated that by reorganizing the neural networks' layers to mimic the structure of the mechanistic model, it is possible to achieve better training rates and prediction accuracy relative to the previously proposed black-box neural networks, while maintaining the interpretability of the mechanistic simulations. Our framework can be used to emulate mechanistic models in a large parameter space and offers an example on the utility of increasing the interpretability of deep learning networks.

Structural basis of HIV-1 maturation inhibitor binding and activity.

HIV-1 maturation inhibitors (MIs), Bevirimat (BVM) and its analogs interfere with the catalytic cleavage of spacer peptide 1 (SP1) from the capsid protein C-terminal domain (CACTD), by binding to and stabilizing the CACTD-SP1 region. MIs are under development as alternative drugs to augment current antiretroviral therapies. Although promising, their mechanism of action and associated virus resistance pathways remain poorly understood at the molecular, biochemical, and structural levels. We report atomic-resolution magic-angle-spinning NMR structures of microcrystalline assemblies of CACTD-SP1 complexed with BVM and/or the assembly cofactor inositol hexakisphosphate (IP6). Our results reveal a mechanism by which BVM disrupts maturation, tightening the 6-helix bundle pore and quenching the motions of SP1 and the simultaneously bound IP6. In addition, BVM-resistant SP1-A1V and SP1-V7A variants exhibit distinct conformational and binding characteristics. Taken together, our study provides a structural explanation for BVM resistance as well as guidance for the design of new MIs.

Recognition of the TDP-43 nuclear localization signal by importin α1/ß.

Cytoplasmic mislocalization of the TAR-DNA binding protein of 43 kDa (TDP-43) leads to large, insoluble aggregates that are a hallmark of amyotrophic lateral sclerosis and frontotemporal dementia. Here, we study how importin α1/ß recognizes TDP-43 bipartite nuclear localization signal (NLS). We find that the NLS makes extensive contacts with importin α1, especially at the minor NLS-binding site. NLS binding results in steric clashes with the C terminus of importin α1 that disrupts the TDP-43 N-terminal domain (NTD) dimerization interface. A putative phosphorylation site in the proximity of TDP-43 R83 at the minor NLS site destabilizes binding to importins by reducing the NLS backbone dynamics. Based on these data, we explain the pathogenic role of several post-translational modifications and mutations in the proximity of TDP-43 minor NLS site that are linked to disease and shed light on the chaperone activity of importin α1/ß.

Toward Convergence in Free Energy Calculations for Protein Conformational Changes: A Case Study on the Thin Gate of Mhp1 Transporter.

It has been challenging to obtain reliable free energies for protein conformational changes from all-atom molecular dynamics simulations, despite the availability of many enhanced sampling techniques. To alleviate the difficulties associated with the enormous complexity of the conformational space, here we propose a few practical strategies for such calculations, including (1) a stringent method to examine convergence by comparing independent simulations starting from different initial coordinates, (2) adoption of multistep schemes in which the complete conformational change consists of multiple transition steps, each sampled using a distinct reaction coordinate, and (3) application of boundary restraints to simplify the conformational space. We demonstrate these strategies on the conformational changes between the outward-facing and outward-occluded states of the Mhp1 membrane transporter, obtaining the equilibrium thermodynamics of the relevant metastable states, the kinetic rates between these states, and the reactive trajectories that reveal the atomic details of spontaneous transitions. Our approaches thus promise convergent and reliable calculations to examine intuition-based hypotheses and to eventually elucidate the underlying molecular mechanisms of reversible conformational changes in complex protein systems.

Thermodynamics of Protein Folding Studied by Umbrella Sampling along a Reaction Coordinate of Native Contacts.

Spontaneous transitions between the native and non-native protein conformations are normally rare events that hardly take place in typical unbiased molecular dynamics simulations. It was recently demonstrated that such transitions can be well described by a reaction coordinate, Q, that represents the collective fraction of the native contacts between the protein atoms. Here we attempt to use this reaction coordinate to enhance the conformational sampling. We perform umbrella sampling simulations with biasing potentials on Q for two model proteins, Trp-Cage and BBA, using the CHARMM force field. Hamiltonian replica exchange is implemented in these simulations to further facilitate the sampling. The simulations appear to have reached satisfactory convergence, resulting in unbiased free energies as a function of Q. In addition to the native structure, multiple folded conformations are identified in the reconstructed equilibrium ensemble. Some conformations without any native contacts nonetheless have rather compact geometries and are stabilized by hydrogen bonds not present in the native structure. Whereas the enhanced sampling along Q reasonably reproduces the equilibrium conformational space, we also find that the folding of an α-helix in Trp-Cage is a slow degree of freedom orthogonal to Q and therefore cannot be accelerated by biasing the reaction coordinate. Overall, we conclude that whereas Q is an excellent parameter to analyze the simulations, it is not necessarily a perfect reaction coordinate for enhanced sampling, and better incorporation of other slow degrees of freedom may further improve this reaction coordinate.