I'm new to bioinformatics, and I have a pretty basic question. Let's say I have a bunch of genes {Xi} and I want to know which one has the most significant on some measurable phenotypic trait Y. Now I have two questions: 1. Assuming that the amount of data is unlimited and I have as many sets of genome: phenotype as I want, can I do some sort of PCA to find out which gene is responsible to the largest part of the variance in Y? Any better Idea? 2. If so, how is the variance in the gene sequence encoded into numbers, so I can do all the variance analysis with it? Thanks in advance, J

  • $\begingroup$ What's the input? Gene expression? SNPs? You need a bit more specific with your question. $\endgroup$
    – Scholar
    Dec 12, 2018 at 17:00
  • $\begingroup$ Let's say the input is a set of SNPs $\endgroup$
    – Julius
    Dec 12, 2018 at 20:27

1 Answer 1


First of, since SNPs as an input are usually encoded as binary (or {0,1,2} when considering heterozygosity) and are very numerous, I'm not sure if the PCA output will be meaningful, but it probably depends on what you're trying to archive.

Second, if your input is a set of SNPs, you're essentially looking at GWAS (genome wide association study) in case the response is categorical, such as case / control, or QTL (quantitative trait loci) analysis, in case your response is continuous, such as gene expression or an individuals height.

GWAS is usually done SNP-wise via hypothesis testing using e.g. Chi-Square or more recently, using machine learning methods such as random forests, because they are (at least theoretically) able to deal with SNP interactions.

QTL analysis can be done using ANOVA, but there might be other methods aswell.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.