New methods of reference class selection applied to genomics data analysis

What is it about?

A local false discovery rate, like any other non-subjective probability, depends on a reference class. The probability that a car weighs over a ton depends on whether the reference class is all cars, all American cars, all sports cars, or all American sports cars. In the same way, the probability that a SNP with certain GERP and polyphen2 scores depends on whether the reference class is all SNPs, all SNPs with similar polyphen2 scores, etc. This type of information is used to obtain more SNP-specific proportions of associated SNPs since the subclass affects the probability of association.

Featured Image

Why is it important?

In empirical Bayes multiple testing, a prior distribution is estimated by considering each gene or other feature corresponding to a null hypothesis to be randomly selected from some reference class of other features. That leads to the problem of selecting a reference class for the application of empirical Bayes methods since each feature belongs to a large number of reference classes.

Perspectives