SAVI: a statistical algorithm for variant frequency identification

Background: Many problems in biomedical research can be posed as a comparison between related samples (healthy vs. disease, subtypes of the same disease, longitudinal data representing the progression of a disease, etc). In the cases in which the distinction has a genetic or epigenetic basis, next-generation sequencing technologies have become a major tool for obtaining the difference between the samples. A commonly occurring application is the identification of somatic mutations occurring in tumor tissue samples driving a single cell to expand clonally. In this case, the progression of the disease can be traced through the trajectory of the frequency of the oncogenic alleles. Thus obtaining precise estimates of the frequency of abnormal alleles at various stages of the disease is paramount to understanding the processes driving it. Although the procedure is conceptually simple, technical difficulties arise due to inhomogeneous samples, existence of competing subclonal populations, and systematic and non-systematic errors introduced by the sequencing technologies. Results: We present a method, Statistical Algorithm for Variant Frequency Identification (SAVI), to estimate the frequency of alleles in a set of samples. The method employs Bayesian analysis and uses an iterative procedure to derive empirical priors. The approach allows for the comparison of allele frequencies across several samples, e. g. normal/tumor pairs and more complex experimental designs comparing multiple samples in tumor progression, as well as analyzing sequencing data from RNA sequencing experiments. Conclusions: Analyzing sequencing data through estimating allele frequencies using empirical Bayes methods is a powerful complement to the ever-increasing throughput of the sequencing technologies.