CIM. n.marcovar, and sample size

[gator....@gmail.com]

Hi Karl,

cim() seems to give a different LOD value for a QTL peak as n.marcovar goes from 3 to 11 when the sample size is low(~100).

Based on your answer here (https://groups.google.com/g/rqtl-disc/c/K1Wnyh5sDBU/m/hZBY5ss7HgAJ) and looking at the code, I am guessing cim() uses forward selection, is there a value in using exhaustive search just for QTL peaks that pass LOD=3 but do not pass the permutation-based threshold, especially when the sample size is low and you want to rescue small effect QTLs?

Prashant

[Karl Broman]

The value of n.marcovar can make a big difference even for much larger sample sizes.

And yes cim() uses forward selection, as mentioned in its help page:

"We first use fill.geno to impute any missing marker genotype data, either via a simple random imputation or using the Viterbi algorithm.

"We then perform forward selection to a fixed number of markers. These will be used (again, with any missing data filled in) as covariates in the subsequent genome scan."

Personally, I don't think the search algorithm is as important as the criterion for comparing models of different sizes. See, for example, Broman and Speed (2002) https://doi.org/10/d4bn8n

karl