BayesFit

Program for fitting analytical models to small-angle scattering data using Bayesian refinement.

a web application is available via GenApp (alpha version): http://bayesfit.genapp.rocks/bayesfit/

*obs: the fortran code is not well documented, so if you wish to use BayesApp, but your model is not implemented, do not hesitate to contact Andreas Larsen for questions or a collaborative project, andreas(dot)larsen(a)bioch(dot)ox(dot)ac(dot)uk.

Citing the program

If you use BayesFit in your work, please cite:

Andreas Haahr Larsen, Lise Arleth, and Steen Hansen (2018). Analysis of small-angle scattering data using model fitting and Bayesian regularization. J Appl Cryst, 51, 1151-1161.

Install and run the program

Operating systems

Only tested on Linux Ubuntu

Installation

Download bayesfit.f and compile it :

gfortran -m64 -O3 bayesfit.f -o bayesfit

OBS, for some systems, bayesapp may be perform better with slightly different compile commands:

gfortran -m64 -O2 bayesfit.f -o bayesfit

gfortran -march=native -O2 bayesfit.f -o bayesfit

Dependencies

fortran
gfortran (for compilation)

Running the program

bayesfit input_file.d

The input file must contain

Line 1: filename with data (headerlines are ignored)
Line 2: model name (describes the function for fitting data, se below)
Line 3: number of steps in function integrals, max iterations in minimization routine at each alpha (latter fixed to 100 in the last minimization step)
Line 4: q_min, q_max (to be included in the fit. to include everything, type, e.g., 0 and 1)
Line 5 and following lines: parameter 1 prior value, parameter 1 prior width, 0 or 1 or 2 (0:fix 1:fit 2:fit with positive constraint)

Example input file (for the for coreshell model):

Isim.dat
coreshell
30 7
0.0 1.0
0.808 0.08 2
4.95e-3 5e-4 1
30.3 3 2
49.5 5 2
101 10 2
3.96 0.4 2

Output

Terminal Output

Column 1: chi^2/M, where M is the number of datapoints Column 2: lambda (used in the Levenberg-Marquardt algorithm)
Column 3: log(alpha)
Column 4: Q
Column 5: dQ (change in Q from previous values - used in convergence criteria)
Column 6: S (the prior value)

For each value of alpha, the best fit is found and the following information is printet
end lambda, iterations refined parameter values and uncertainties
prior parameter values and prior uncertainties
final chi^2_r, -log(evidence), and Ng

At the end the following is printed: optimal alpha and evidence at that value computation time

Output files

parameters.d

prior and refined parameter values, as well as Chi-square, reduced Chi-square, final alpha, number of good parameters (Ng), number of fitted datapoints (M), evidence, value of constraint (S), and value of alpha*S.

prior.d

Intensity calculated with the prior values, which are also the initial values in the fitting algorithm.

fit.d

Most probable fit after model refinement for all alpha-values.

data.d

Data used in the fitting process (no headerlines

Models and Example data

Four models have been implemented so far in BayesFit.
The models can be run with the python3 script run_examples.py. Edit the line "i = 0" to select what example model to run, and run the script:

python run_examples.py

Nanodisc

Name of model (for input file): nanodisc
Name of function (in BayesFit.f): fct_nano
Description: Elliptical nanodiscs with a purification tag. The disc is build up of cylinder form factors and the tag is modelled as a random coil.
Reference: Andreas Haahr Larsen, Lise Arleth, and Steen Hansen (2018). Analysis of small-angle scattering data using model fitting and Bayesian regularization. J Appl Cryst, 51, 1151-1161.
Example data: Dataset5.rad
Example input file: input_nano.d
parameters:

Background [1/cm]
Concentration [uM]
Volume, lipid [A^3]
Volume, lipid tail [A^3]
Correction of volume, protein
Number of lipids per disc
Thickness of protein belt [A]
Roughness [A^2]
Area per lipid headgroup [A^2]
Ellipticity of lipid core
Number of waters per lipid
Radius of gyration of tag [A]

Micelle

Name of model (for input file): micelle
Name of function (in BayesFit.f): fct_mic
Description: Core-shell detergent micelle, DDM micelles, concentration 30 mM.
Reference: Andreas Haahr Larsen, Lise Arleth, and Steen Hansen (2018). Analysis of small-angle scattering data using model fitting and Bayesian regularization. J Appl Cryst, 51, 1151-1161.
Example data: RB_DDM30mM.dat
Example input file: input_micelle.d
parameters:

Concentration [mM]
Background [1/cm]
Aggregation number, N
Contrast shell [e/A^3]*
Contrast core [e/A^3 ]*
Roughness, sigma [A]
Ellipticity, epsilon
e is the scattering lenght density of an electron

CoreShell

Name of model (for input file): coreshell
Name of cunction (in BayesFit.f): fct_coreshell
Description: Idealised spherical core-shell particle.
Reference:
Example data: Isim.dat
Example input file: input_coreshell_good_prior.d, input_coreshell_poor_prior.d
parameters:

I(0) [1/cm]
Constant backbround [1/cm]
Radius of core [A]
Radius of shell [A]
Excess scattering length density (contrast) core [arb. units]
Excess scattering length density (contrast) shell [arb. units]

Test model of spheres

Name of model (for input file): test
Name of function (in BayesFit.f): fct_test
Description: Simple sphere with background.

Versions

version 1

First release, August 2018

version 2

Release September 2020
Cleaned up output
Added model CoreShell

License

BayesFit is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

BayesFit is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details (http://www.gnu.org/licenses/).

Acknowledgements

CoNeXT, the Carlsberg Foundation, and University of Copenhagen for co-funding the project.

Suggestions for future development

include resolution effects.
option for inclusion of several datasets.
restructure, so models are independent modules.
expand with more models.

niels-bohr-institute-xns-structbiophys / bayesfit Goto Github PK

bayesfit's Introduction