I have different conformations, created by MDs, representing a protein. I have compared them with SAXS data by using Crysol. some of Chi square values obtained from Crysol are 0.04, 0.036, 0.034, 0.03, 0.025. In this case, which conformation is the best to represent the protein structure: the one that gives closest chi square to 1 or the conformation that gives the lowest chi square?
Dear Omar,
in case of SAXS Chi^2 depends on the difference between two curve and also on the experimental error.
If Chi^2 is larger than 1 than the two curves are different, while smaller than one, the errors are too high.
Target value is 1.
You might check the following presentation by D. Franke on this topic.
https://www.embl-hamburg.de/biosaxs/courses/embo2018/slides/franke-statistics.pdf
Best,
Martin
Thank you guys for sharing your knowledge with me.
Depending on what Martin has mentioned above and the provided file, fitting a model to my data gives high statistical errors (very low Chi^2). I am facing this low value not only with Crysol but also with GNOM even though curves showing a good fit (attachment).
Here my questions:
- Is it reliable to use low Chi^2 values to decide which conformation is close to the experimental one, in case of Crysol and MDs or to create an envelop, in case of GNOM? If not what is the alternatives?
- What is the reason of getting low Chi^2?
Please if you have any reference, do not hesitate to share it.
Best regards,
Omar
Hi Omar,
indeed the low Chi^2 values points to issues of the (experimental) error of your data (These are experimental data, right?). This is unfortunately not seldom. However, there are others tools available to determine the quality of the SAS curve of a model with respect to the experimental data -WITHOUT need of the error.
One is the Correlation Maps approach, which is now standardly used in the ATSAS software: Either using in the PrimusQt GUI Processing -> Data Comparison. OR in command line datcmp.
More information on this method can be found in the respective paper
https://www.nature.com/articles/nmeth.3358
You might also check the SAXS forum for additional information:
https://www.saxier.org/forum/
Cheers,
Martin
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biomolecules
- Peptides