Federated learning for multi-omics: A performance evaluation in Parkinson's disease.

While machine learning (ML) research has recently grown more in popularity, its application in the omics domain is constrained by access to sufficiently large, high-quality datasets needed to train ML models. Federated learning (FL) represents an opportunity to enable collaborative curation of such datasets among participating institutions. We compare the simulated performance of several models trained using FL against classically trained ML models on the task of multi-omics Parkinson's disease prediction. We find that FL model performance tracks centrally trained ML models, where the most performant FL model achieves an AUC-PR of 0.876 ± 0.009, 0.014 ± 0.003 less than its centrally trained variation. We also determine that the dispersion of samples within a federation plays a meaningful role in model performance. Our study implements several open-source FL frameworks and aims to highlight some of the challenges and opportunities when applying these collaborative methods in multi-omics studies.

Federated Learning for multi-omics: a performance evaluation in Parkinson's disease.

While machine learning (ML) research has recently grown more in popularity, its application in the omics domain is constrained by access to sufficiently large, high-quality datasets needed to train ML models. Federated Learning (FL) represents an opportunity to enable collaborative curation of such datasets among participating institutions. We compare the simulated performance of several models trained using FL against classically trained ML models on the task of multi-omics Parkinson's Disease prediction. We find that FL model performance tracks centrally trained ML models, where the most performant FL model achieves an AUC-PR of 0.876 ± 0.009, 0.014 ± 0.003 less than its centrally trained variation. We also determine that the dispersion of samples within a federation plays a meaningful role in model performance. Our study implements several open source FL frameworks and aims to highlight some of the challenges and opportunities when applying these collaborative methods in multi-omics studies.

Insights into autoregulation of a membrane protein complex by its cytoplasmic domains.

Membrane transporters mediate the passage of molecules across membranes and are essential for cellular function. While the transmembrane region of these proteins is responsible for substrate transport, often the cytoplasmic regions are required for modulating their activity. However, it can be difficult to obtain atomic-resolution descriptions of these autoregulatory domains by classical structural biology techniques, especially if they lack a single, defined structure. The betaine permease, BetP, a homotrimer, is a prominent and well-studied example of a membrane protein whose autoregulation depends on cytoplasmic N- and C-terminal segments. These domains sense and transduce changes in K+ concentration and in lipid bilayer properties caused by osmotic stress. However, structural data for these terminal domains is incomplete, which hinders a clear description of the molecular mechanism of autoregulation. Here we used microsecond-scale molecular simulations of the BetP trimer to compare reported conformations of the 45-amino-acid long C-terminal tails. The simulations provide support for the idea that the conformation derived from electron microscopy (EM) data represents a more stable global orientation of the C-terminal segment under downregulating conditions while also providing a detailed molecular description of its dynamics and highlighting specific interactions with lipids, ions, and neighboring transporter subunits. A missing piece of the molecular puzzle is the N-terminal segment, whose dynamic nature has prevented structural characterization. Using Rosetta to generate ensembles of de novo conformations in the context of the EM-derived structure robustly identifies two features of the N-terminal tail, namely 1) short helical elements and 2) an orientation that would confine potential interactions to the protomer in the counterclockwise direction (viewed from the cytoplasm). Since each C-terminal tail only contacts the protomer in the clockwise direction, these results indicate an intricate interplay between the three protomers of BetP in the downregulated protein and a multidirectionality that may facilitate autoregulation of transport.

Interpreting Hydrogen-Deuterium Exchange Experiments with Molecular Simulations: Tutorials and Applications of the HDXer Ensemble Reweighting Software [Article v1.0].

Hydrogen-deuterium exchange (HDX) is a comprehensive yet detailed probe of protein structure and dynamics and, coupled to mass spectrometry, has become a powerful tool for investigating an increasingly large array of systems. Computer simulations are often used to help rationalize experimental observations of exchange, but interpretations have frequently been limited to simple, subjective correlations between microscopic dynamical fluctuations and the observed macroscopic exchange behavior. With this in mind, we previously developed the HDX ensemble reweighting approach and associated software, HDXer, to aid the objective interpretation of HDX data using molecular simulations. HDXer has two main functions; first, to compute H-D exchange rates that describe each structure in a candidate ensemble of protein structures, for example from molecular simulations, and second, to objectively reweight the conformational populations present in a candidate ensemble to conform to experimental exchange data. In this article, we first describe the HDXer approach, theory, and implementation. We then guide users through a suite of tutorials that demonstrate the practical aspects of preparing experimental data, computing HDX levels from molecular simulations, and performing ensemble reweighting analyses. Finally we provide a practical discussion of the capabilities and limitations of the HDXer methods including recommendations for a user's own analyses. Overall, this article is intended to provide an up-to-date, pedagogical counterpart to the software, which is freely available at https://github.com/Lucy-Forrest-Lab/HDXer.