excuse me for that as recently joined this field and trying to make my way. I have a data set of 1545 SNP markers from NGS for many traits of 63 individuals (F2 intercross)
the genotypes analyzed with stacks and I constructed linkage map with Onemap software.
when I tried to perform QTL analysis in R, The results of scanone and scanetwo show loci with high LOD but no loci with peaks higher than the significant threshold which obtained from permutation test. so no pair of loci meeting the threshold at any leve
I tried at 5,20 and even 60% with no use always no pair of loci meeting the criteria suggested by the test. i need to understand what the matter and how can I solve this.
regards.
Significance tests are very important to avoid the detection of false positives signals in your data. Sample size, precise phenotyping, phenotypic distribution and genomic coverage of markers are very important to associate any phenotype to genotype and to say as QTL.
Check the normality of phenotypic distributions. Try to apply transformations to data, if its is skewed. However, in your case, sample size seems to be very small, which is probably the culprit.
With small sample size, your data will become too much skewed.
Significance tests are very important to avoid the detection of false positives signals in your data. Sample size, precise phenotyping, phenotypic distribution and genomic coverage of markers are very important to associate any phenotype to genotype and to say as QTL.
Check the normality of phenotypic distributions. Try to apply transformations to data, if its is skewed. However, in your case, sample size seems to be very small, which is probably the culprit.
With small sample size, your data will become too much skewed.
In general，trait values do not follow a normal distribution, because they are caused by multiple causal loci and various other factors. They should follow some mixed distribution.
Model miss-specification would be a key reason, especially for small sample size. Another, permutation based p-value may not be accurate enough even if the adopted working model well fits the true data distribution.
If you have a few factors of large effect segregating in the F2 and a decent trait difference in the F0 it should always reveal a peak if the gene does not interact with the background or environment (it's a real deal). Generation 0 has to have a nice difference in the trait of interest between the parents which needs to be distribution-appropriate to the model. Check your residual Q-Q plot to see. In the G1 you might want to replicate the G1 crosses if your original G0 lines are not highly inbred lines. Significant results can have small phenotypic effect sizes - which is not really the goal is it? A very significant effect of nothing? Large effect sizes with slightly smaller q-values (multiply corrected however you want - I like benjamini hochberg FDR not permutation tests or bonferroni) can still be interesting at least from a commercial agricultural point of view. Most heritabilities are not statistically significant for example. So you might want to investigate a bivariate plot of your effect size and your q-values. Think about interaction with the genetic background e.g. over a few F2's of different origins and then maybe over F1 replicate crosses (partially inbred/ LD flavors) might be useful depending on the design. You might get a good heterogeneity P-value but not really an exceptionally grand total or marginal P value or likelihood - making the mistake of interpreting interaction as noise that cancels. This can still be interesting highlighting significant interaction ( not stochastic sampling) so there are models to incorporate "the other QTL" such as multiple interval mapping - i.e. not just point or interval but using the intervals of other loci as cofactors etc. Do you? This is all dependant on the trait etiology and the design. If a locus winds up highly significant in one cross and not another I would still go after interactions with background and more crosses fixing the original genotype and interaction system as much as you can. Not knowing the details this is what I would offer. Cheers and good luck!
Daniel
- Badges
- Science method