Replic2 – software for evaluating replication datasets
Jurg Ott 8 Feb 2025
Scenario
Assume you have two independent case-control datasets, e.g. males and females, that have been genotyped for M1 and M2 variants (SNPs), respectively. You select a subset of these, N1 and N2, as being particularly promising, for example, because they are significant or because they show odds ratios exceeding 10 each. Upon combining N1 and N2 into a single dataset (spreadsheet), you find that N12 of these variants are common between the two groups. How likely is this occurring just by chance, or can you conclude that group 2 represents a significant replication for group 1 by virtue of these N12 common variants?
Replic2 program
This question can be answered by simulating the following null situation: You randomly pick N1 variants out of a larger set of M1 variants for group 1 and analogously for group 2, then combine them to see how many random matches you find, N12rand. You do this many times to obtain, for example, 100,000 such N12rand numbers. The empirical significance level, p, is then given by the proportion of N12rand values equal to or exceeding the observed number N12 of matches.
The easiest way to use the Replic2 program is to run it in a command box, either in Windows or Linux, followed on the command line by the name of an input file. For example, you type Replic2 sample.in, where the sample.in file contains the following lines:
1000 1000 Two large sample sizes 100 100 Two sets of SNPs selected from the large sample sizes 100000 Number of replicates for p-value calculation 3 Number of SNPs N12, common to the two groups... 5 ... repeat as often as desired 8 10 20 30 50 -1 Finish with "-1" or just end the input
The resulting output file will then list each number N12 of common SNPs and their associated p-values. Because we take the observed data as one of the null datasets, there will never be a zero p-value. For example, with 100,000 replicates, the smallest possible significance level is p = 0.00001, which may be interpreted as p < 0.00001.
The program currently works only for a relatively small number of variants. As time permits, I will modify the text (Pascal script) to accommodate larger numbers. The program is available on github.
