Experimental noise in small-angle scattering can be assessed using the Bayesian indirect Fourier transformation

TL;DR

This paper introduces a Bayesian indirect Fourier transformation method to evaluate and correct experimental errors in small-angle scattering data, improving model fitting accuracy and error assessment.

Contribution

It presents a novel Bayesian approach for assessing and rescaling experimental errors in small-angle scattering data, applicable to both simulated and experimental datasets.

Findings

Effective in detecting over- or under-estimated errors

Can rescale errors for better model fitting

Helps determine appropriate reduced χ² targets

Abstract

Small-angle X-ray and neutron scattering are widely used to investigate soft matter and biophysical systems. The experimental errors are essential when assessing how well a hypothesized model fits the data. Likewise, they are important when weights are assigned to multiple data sets used to refine the same model. Therefore, it is problematic when experimental errors are over- or under-estimated. A method is presented, using Bayesian indirect Fourier transformation for small-angle scattering data, to assess whether or not a given small-angle scattering data set has over- or under-estimated experimental errors. The method is effective on both simulated and experimental data, and can be used to assess and rescale the errors accordingly. Even if the estimated experimental errors are appropriate, it is ambiguous whether or not a model fits sufficiently well, as the "true" reduced $\chi^2$ of the data is not necessarily unity. This is particularly relevant for approaches where overfitting is an inherent challenge, such as reweighting of a simulated molecular dynamics trajectory against small-angle scattering data or ab initio modelling. Using the outlined method, it is shown that one can determine what reduced $\chi^2$ to aim for when fitting a model against small-angle scattering data. The method is easily accessible via the web interface BayesApp.