Better GE rate simulation?

[jamonterotena]

Hello again,

I noticed that when simulating genotyping errors the actual genotyping error rate can differ from the input. This is because the input GE rate is taken as a probability that any site suffers a GE:

markers = which(rbinom(length(masked_genotypes), 1, error_rate) == 1)

I guess this is a matter of the definition of genotyping error rate and not a bug, but I think it's more intuitive to think that GE rate 6% means that there are 6 genotyping errors in 100 sites, and not maybe 4, 7 or 6.

I suggest something like the following

total_GEs = error_rate*length(masked_genotypes)
markers = sample(1:length(masked_genotypes), total_GEs , replace = FALSE) ## Randomly pick a proportion of the sites equal to error_rate for GEs

[gregorgorjanc]

@jamonterotena what/which piece of code in AlphaSimR are you referring to?

[jamonterotena]

@gregorgorjanc recombinations. Not sure of the name of the script, sorry.

[gregorgorjanc]

@jamonterotena are you referring to a function, if so, which one? If you are referring to a script, which script are you referring to?

If this is more of a general discussion, then https://github.com/gaynorr/AlphaSimR/discussions might be a better place. If I am getting your question right, then it really depends what you want to do. Saying that genotyping error rate is 6% and getting, say, 5, errors out of a 100 genotypes, is just how probability works. If you want exactly 6 errors, then you can use your solution, but note that in reality we always have some deviations/distribution of values;)

[gaynorr]

@jamonterotena I changed this to a discussion topic, because I don't support genotyping error within AlphaSimR itself. However, a discussion about adding this to AlphaSimR makes sense. I've never added it to AlphaSimR for two reasons.

First, I'm not sure if there is a single method for representing genotyping error that makes sense for multiple use cases. The method that I tend to use myself, is to identify an individual and a locus with a genotyping error and then shift the genotype by a score of 1. In the diploid case, this means a homozygote gets reported as a heterozygote and a heterozygote randomly shifts to one of the homozygotes.

The second reasons I haven't implemented anything is because there isn't an easy way to store these errors within an AlphaSimR population.

[jamonterotena]

The method you applied in the script linked below assumes constant genotyping error rate by SNP and individual. This ignores the effect of sample contamination or 'bad' SNPs.

Just out of curiosity @gaynorr @gregorgorjanc, did you find/developed any way to consider this when adding errors?

[gregorgorjanc]

@jamonterotena you could sample the rate from a beta distribution for each SNP and for each individual, which will give you the variation you might be after. Maybe you would then multiply/add the two rates to get SNP-specific rate of each SNP of each individual.

[jamonterotena]

This is the script I'm referring to, line 101:
https://github.com/gaynorr/AlphaSimR_Examples/blob/master/recombination_simulations/simulate_chromosome.r

I'm using it to compare the precision of a crossover-calling program at different GE rates so it works better for me if the input GE rates are actually the proportion of false GEs in the final genotypes, but I understand it's also senseful to take GE rate as the probability of any site to be wrong.