VNTR prediction on sequence characteristics using long-read annotation and validation by short-read pileup

Tandem repeats (TRs) are contiguously repetitive sequences with a high mutation rate. Several human diseases have been associated with an expansion of TR, a mutation which constitutes a change in their number of repetitions. Nevertheless, these Variable Number Tandem Repeats (VNTRs) have not been included in many genome-wide studies. The reason is that VNTR genotyping is inaccurate using short-read sequencing while new technology like long-read sequencing is expensive and lacks throughput. Here, we propose a sequence based random forest classifier that is able to predict variable expansion of TR regions, given by incomplete VNTR annotation from long-read sequencing of 5 haplotypes. The classifier mainly predicted VNTRs using the features TR length. The second most used feature is a novel finding: the Mfold predicted likelihood of self-folding for which more stable foldings are correlated with VNTRs. We validated VNTR candidates predicted by this classifier by clustering short-read pileup patterns compared across 17 genomes. TRs labeled VNTR by the classifier showed similar local variance in their pileup profiles. Contactdiederik.cvb@gmail.com Supplementary informationSupplementary data are available at bioRxiv