A Functional Bayesian Model for Hydrogen–Deuterium Exchange Mass Spectrometry

Oliver M. Crook, Nathan Gittens, Chun‐wa Chung, Charlotte M. DeaneJournal of Proteome Research 2023 · DOI

TL;DR

This paper introduces a flexible Bayesian framework for analyzing Hydrogen-Deuterium Exchange Mass Spectrometry (HDX-MS) data. This new approach offers improved algorithmic stability, robust uncertainty quantification, and the ability to derive statistical insights previously inaccessible, leading to more sensitive and interpretable results in protein interaction studies.

What Problem Does This Paper Address?

Proteins are dynamic molecules that change their structure in response to binding partners or environmental shifts. Hydrogen-Deuterium Exchange Mass Spectrometry (HDX-MS) is a powerful technique to study these conformational changes by tracking how quickly deuterium atoms replace hydrogen atoms in a protein when incubated in heavy water. This exchange rate is context-dependent and can reveal subtle structural alterations. While incorporating the time course of deuterium incorporation into statistical analysis has improved interpretation, existing methods often rely on restrictive assumptions about the exchange kinetics, limiting their flexibility and power. This paper aims to overcome these limitations by developing a more adaptable and statistically rigorous Bayesian framework for HDX-MS data analysis.

Methods

The core of this work is the development of a functional Bayesian model for HDX-MS data. Instead of relying on fixed kinetic models (like logistic or Weibull), the Bayesian approach allows for flexible modeling of time-dependent deuterium incorporation. Key methodological contributions include:

  • Bayesian Inference Framework: Reframing HDX-MS analysis within a Bayesian framework. This allows for the incorporation of prior knowledge, shrinkage of residuals, and regularization of model parameters, leading to more stable inference, especially with complex data or limited observations.
  • Uncertainty Quantification: The Bayesian approach naturally provides probability distributions for model parameters and derived quantities, enabling robust confidence reporting and a deeper understanding of the reliability of findings.
  • Model Selection: The framework facilitates rigorous model selection, allowing researchers to objectively choose the best kinetic model (e.g., logistic vs. Weibull) for their data, particularly important when dealing with complex exchange mechanisms like EX1 dynamics.
  • Bridge Sampling: A computational technique used to approximate marginal likelihoods, which are crucial for Bayesian model comparison.
  • Prior and Posterior Predictive Checks: Methods to assess the quality and validity of the generative and predictive models.
  • Markov Chain Monte Carlo (MCMC) with NUTS: The implementation uses the No-U-Turn Sampler (NUTS), a variant of Hamiltonian Monte Carlo, for efficient and robust parameter inference.

Datasets

The paper utilizes several datasets for evaluation and demonstration:

  • Spike-in HDX-MS experiment: Used to demonstrate rigorous model selection capabilities.
  • Epitope mapping experiment (HOIP-RBR complexed with dAb25 and dAb3): Used to showcase improved interpretation and identification of binding interfaces.
  • Small molecule case-study (BRD4 complexed with I-BET151): Used to demonstrate increased sensitivity and the ability to perform analysis with reduced data.
  • Maltose-Binding Protein (MBP) permutation experiment: Used to demonstrate control over false discoveries.

No new benchmark datasets were introduced; instead, existing experimental data were re-analyzed using the novel Bayesian approach.

Key Insights and Ideas

  • Flexibility is Key: The Bayesian framework's inherent flexibility allows it to handle complex HDX-MS data that might confound simpler, assumption-laden models. This is crucial for understanding intricate protein-ligand interactions.
  • Uncertainty is Informative: Quantifying uncertainty isn't just about reporting error bars; it's about understanding the confidence in specific findings. The Bayesian approach provides rich probabilistic outputs that can be directly interpreted.
  • Beyond Simple Differences: The ability to calculate probabilities of deuterium differences exceeding specific thresholds (e.g., P(ΔD > d)) provides a more nuanced understanding of the magnitude and significance of observed changes.
  • Model Selection Matters: Objectively selecting the correct kinetic model (logistic vs. Weibull) is vital for accurate interpretation, especially when EX1 dynamics are present. The Bayesian approach automates this process.
  • Bridging Structural and Dynamic Data: The Bayesian HDX-MS analysis can provide complementary evidence to structural data (like crystal structures), helping to sublocalize interaction sites and validate findings.

Key Results

  • Improved Sensitivity: The Bayesian approach demonstrated increased sensitivity in detecting deuterium incorporation changes, particularly in the BRD4-I-BET151 case study, where it identified a higher percentage of protected peptides compared to a previous empirical Bayes method.
  • Accurate Model Selection: The method successfully identified Weibull kinetics as preferred over logistic models for data exhibiting EX1 dynamics in a spike-in experiment.
  • Enhanced Epitope Mapping: In the HOIP-RBR epitope mapping experiments, the Bayesian approach identified more potential binding interfaces and provided stronger evidence for these sites compared to previous methods. For HOIP-RBR-dAb3, it identified three potential interfaces consistent with crystal structure data, where previous methods failed to detect any epitopes.
  • Robust False Positive Control: A permutation experiment on MBP data showed excellent control of error rates, with posterior probabilities rarely exceeding 0.05, indicating a low rate of false positives.
  • Good Calibration: The Brier scores for the Bayesian approach were consistently close to zero, indicating high-quality calibration of posterior probabilities.
  • Reduced Data Requirements: The Bayesian approach showed remarkable resilience to data ablation, maintaining performance with significantly fewer data points than the empirical Bayes method.

Limitations and Future Work

  • Computational Cost: While NUTS is efficient, Bayesian inference can still be computationally intensive, especially for very large datasets or complex models.
  • Prior Specification: The choice of priors can influence results, and while the paper discusses prior predictive checks, careful consideration is still needed.

The authors suggest that this Bayesian framework can aid in the rapid profiling of small molecule inhibitor modes of action and can be integrated with machine learning approaches for further insights. The flexibility and quantitative power of the method open doors for more sophisticated analyses of complex HDX-MS datasets.

Summary updated

Abstract

Proteins often undergo structural perturbations upon binding to other proteins or ligands or when they are subjected to environmental changes. Hydrogen-deuterium exchange mass spectrometry (HDX-MS) can be used to explore conformational changes in proteins by examining differences in the rate of deuterium incorporation in different contexts. To determine deuterium incorporation rates, HDX-MS measurements are typically made over a time course. Recently introduced methods show that incorporating the temporal dimension into the statistical analysis improves power and interpretation. However, these approaches have technical assumptions that hinder their flexibility. Here, we propose a more flexible methodology by reframing these methods in a Bayesian framework. Our proposed framework has improved algorithmic stability, allows us to perform uncertainty quantification, and can calculate statistical quantities that are inaccessible to other approaches. We demonstrate the general applicability of the method by showing it can perform rigorous model selection on a spike-in HDX-MS experiment, improved interpretation in an epitope mapping experiment, and increased sensitivity in a small molecule case-study. Bayesian analysis of an HDX experiment with an antibody dimer bound to an E3 ubiquitin ligase identifies at least two interaction interfaces where previous methods obtained confounding results due to the complexities of conformational changes on binding. Our findings are consistent with the cocrystal structure of these proteins, demonstrating a bayesian approach can identify important binding epitopes from HDX data. We also generate HDX-MS data of the bromodomain-containing protein BRD4 in complex with GSK1210151A to demonstrate the increased sensitivity of adopting a Bayesian approach.
BibTeX
@article{crook2023functional,
  author = {Oliver M. Crook and Nathan Gittens and Chun‐wa Chung and Charlotte M. Deane},
  title = {{A Functional Bayesian Model for Hydrogen–Deuterium Exchange Mass Spectrometry}},
  journal = {Journal of Proteome Research},
  year = {2023},
  doi = {10.1021/acs.jproteome.3c00297},
}
Figure