Overview
This research presents a framework for inferring interaction potentials from cross-linking mass spectrometry (XL-MS) data. The primary objective is to adapt XL-MS information, which provides insights into pairwise spatial proximity of proteins through inter-molecular linkers, for coarse-grained interacting particle models. The framework addresses limitations in existing methods, particularly concerning the unique inversion of observables like coordination numbers, which are functionals of radial distribution functions (RDFs).
Research Context
Cross-linking mass spectrometry (XL-MS) serves as a quantitative technique for investigating intra-protein structural information and protein-protein interactions. It yields data on the pairwise spatial proximity of proteins in a system. While XL-MS provides valuable information, systematic approaches for utilizing this data to parameterize coarse-grained interacting particle models have been limited. A common approach involves directly fitting radial distribution functions (RDFs). However, relying solely on RDF fitting overlooks numerous observables, such as coordination numbers, which are derived from RDFs and cannot be uniquely inverted using direct fitting methods. The developed framework aims to overcome this limitation, particularly in contexts involving potentially phase-separated mixtures, which are relevant to certain XL-MS results.
Approach
The developed framework establishes a connection between the problem of parameterizing interaction potentials from XL-MS observables and the inverse Henderson problem. It adapts existing algorithms for its numerical solution. Specifically, Iterative Boltzmann Inversion and Iterative Monte Carlo algorithms were adapted for this purpose. The researchers derived both exact and low-density limit gradient approximations.
Two new algorithms were proposed, based on an adaptation of the predictor-corrector framework. In total, several optimization algorithms were evaluated using biologically realistic ten-component test systems. The evaluation aimed to assess their efficiency and accuracy in recovering interaction parameters.
Findings
- For homogeneous fluids, all evaluated methods demonstrated exceptional efficiency and accuracy in parameterizing interaction potentials.
- In a challenging three-phase system, three specific algorithms successfully recovered the correct parameters. These algorithms include:
- Adam, employing the low-density derivative.
- Gradient descent, also employing the low-density derivative.
- Newton's method, utilizing the exact gradient.
Why This Matters
The results establish a pathway for transitioning from XL-MS experimental data to coarse-grained protein models in systems where phase separation influences biological function. This could facilitate new investigations into biomolecular condensates and protein aggregation.