Model Selection Using Replica Averaging with Bayesian Inference of Conformational Populations

Area of Science:

• Computational Chemistry
• Biophysics
• Molecular Dynamics

Background:

• Simulated molecular ensembles often require reconciliation with sparse or noisy experimental data.
• Existing reweighting algorithms may not fully capture uncertainties or provide objective model selection metrics.

Purpose of the Study:

• To introduce an enhanced Bayesian Inference of Conformational Populations (BICePs) algorithm for reconciling simulated ensembles with experimental data.
• To develop a robust method for sampling posterior distributions of conformational populations and assessing force field performance.

Main Methods:

• Modified BICePs algorithm incorporating replica-averaging in its forward model.
• Application to reweighting conformational ensembles of the mini-protein chignolin using extensive experimental data (NOE, chemical shifts, J-couplings).
• Utilized the BICePs score, a free energy-like quantity, for objective model selection and force field evaluation.

Main Results:

• Reweighted conformational populations consistently favored the correctly folded chignolin structure across nine tested force fields.
• The BICePs score provided a reliable metric for evaluating force field performance, aligning with previous studies.
• The algorithm effectively handled uncertainties and outliers in experimental data.

Conclusions:

• The enhanced BICePs algorithm offers significant advantages for ensemble reweighting and model selection in molecular simulations.
• BICePs provides a powerful tool for assessing force field accuracy and improving the interpretation of experimental data.
• This approach holds promise for future applications in computational biophysics and drug discovery.