How to select MSI loci most predictive of a sample's status ??

[lidd77]

Hi to you ,
I am new to MSI and MANTIS .

I have had a BED file with recording MS loci , but I wanna reduce the number of this MS loci for the bigger accuracy of MANTIS ,because MANTIS would be affected by the loci number .

I have read MANTIS paper , you said like this ,"To assess the effect of considering different
numbers and selective microsatellite loci on MSI analysis, we identified the 10, 20, 30, 40, 50, 100, 250,500 and 1000 loci most predictive of a sample’s status across COAD/READ, UCEC and STAD cohorts, for mSINGS, MSISensor and MANTIS ",
so could you tell me the detailed introduction about how to select the most predictive MSI loci ?
what alorgthims did you use ?

I heard some people just used 22 MSI loci to evaluate their tumor samples , but I don't know how they select these 22 loci , do you know this ?

Best

[rbonneville]

We selected the most predictive loci in the MANTIS paper with reference to "gold-standard" MSI-PCR calls available in the TCGA clinical datasets, by selecting loci with the highest scores in known unstable cases and lowest in known stable cases. If you have data from known MSI-H and MSS samples you could do this with your own data, but if not, what sort of capture are you using (whole exome, targeted, etc.)?

[lidd77]

@rbonneville
It's targeted capture that I'm using .
I also have data with known MSI-H and MSS samples ,but the number of MSI-H samples is not many, only totally 5 samples, the MSS samples number is 10 .
About what you said, 'by selecting loci with the highest scores in known unstable cases and lowest in known stable cases' , you mean that firstly I should use MANTIS to run all my paired tumor-normal samples, then I can use MANTIS scores to select loci ? .
How many loci should I select for the higheset scores and the loweset in my known unstable and stable cases ?