Dear colleagues,
I am a little bit curious about your favorite TM-score cutoff when you are comparing two protein structures. Is there some "consensus" analogical to a p-value of 0.05? But yeah, someone prefers to use a p-value of 0.01, or even 0.001 ... To be clear, I am preparing a publication accompanying also structural similarity searches and am not sure which TM-score cutoff to choose.
Definition: TM-score has the value in (0,1], where 1 indicates a perfect match between two structures.
So can I choose e.g. cutoff of 0.55 or 0.56, which gives me a meaningful number of results (similar proteins to my query PDB structure) that are interesting from the biological point of view? Because when I choose a slightly nicer threshold of 0.5, the number of results/structures rise exponentially :D
Thank you for your suggestion
Martin
Dear Martin Bartas
This paper is helpful
https://academic.oup.com/bioinformatics/article/34/10/1719/4769500?login=true
First of all, have you redundancy reduced your search space by sequence similarity (e.g. 99% or 90%)? Otherwise, you might be comparing different structures that were obtained for the same protein using a slightly different experimental/data processing setup.
To answer your question, you can plot the number of sequences you get versus
TM cutoff applied, then select 2-3 cutoffs based on the need for having a "reasonable number" and discuss whether the results/conclusions remain robust for these datasets. As long as you are transparent about how you performed your analysis and show that your conclusions are robust regardless of the exact cutoff chosen I don't think anyone will be too critical!
Dear Asadollah Asadi , thank you for a useful paper
Dear Jan Zaucha , thank you for your suggestion, I work with a set of representative protein structures (1 structure for 1 protein sequence). Regarding TM cutoff, thank you for your optimistic view, hopefully, reviewers will be OK with my "customized" cutoff of TM = 0.55 :)
Best wishes,
Martin