Hi everybody,
Please could you tell me the best filtering cutoff (specially for MAF and missing data) I should apply to my SNPs called from GBS data before doing population structure analysis in my species. My populations are landrace and have not reference genome.
Thanks in advance!
Hi Jacky --
I have recently been doing some GBS data quality control on plant landraces as well. It really depends on your dataset and what you feel is a good balance between missing data and eliminating SNPs and/or samples. I prefer a more conservative approach. In one of my studies, we removed SNPs using VCFtools such that if a SNP had more than 20% missing across taxa, it would be removed. In another study, we used a more conservative measure - cutting out SNPs that had more than 10% missing across taxa. After all this, I think that a cutoff value of SNPs that are missing no more than 20% is a good trade off.
Hope this helps!
Ashley
Unfortunately there is no out-of-the-box answer, but just to give you some pointers from my experience (mainly genomic selection, but also GWAS):
- missing rate per marker: I do trials with thresholds in the 10-50% range, but usually 20% or 30% are where the maximum predictive ability (for GS) is found
- missing rate per sample: please consider this one too! Some samples will have very high missing rates. I filter the same rate as per marker
As a note, I do filtering in two steps: first remove the ugly stuff (markers and sample with, say, more than 80% missing rate). Only then perform the actual filtering at, say, 30%. The reason for doing this is that you don't want all the dirty data (especially very sparse, unusable markers) to affect the final cleaning process.
- MAF: unless you have really good reasons, remove markers with very low MAF. I usually do 2.5%, but anything between 2% and 5% will probably be ok
- imputation algorithm: I did a methodological study (attached) on rice and alfalfa, and it turns out that the very best (but slow) is Random Forest Imputation. Close second best (and waaay faster) is K-nearest Neighbours imputation, which I usually use
Article Marker imputation efficiency for genotyping-by-sequencing da...
Hello,
i am looking for GBS data, mapped and not filtered, could any one help ?
cheers
- Badges
- Science topic
- Similar topics
- Agricultural Science
- Agronomy