Text

theses

Genome sequencing is becoming an indispensable part of biological research. Mutations identified in genomic sequence contribute to explanations of disease, phenotypic variation, and evolutionary adaptation. Increasing reliance on next generation sequencing (NGS) data necessitates efficient and accurate means of genome analysis. We developed two algorithms, GROM-RD and GROM, to address current issues of mutation calling in NGS data. GROM-RD analyzes multiple biases in read coverage to improve copy number variation (CNV) detection in NGS data. GROM-RD takes a two-tiered approach to complex and repetitive segments, while incorporating excessive coverage masking, GC weighting, GS bias normalization, dinucleotide repeat bias normalization, and a sliding-window break-point calculator. Current NGS projects produce massive amounts of data, often on multiple samples; with several approaches designed specifically for each variant, use of multiple algorithms is necessary. GROM provides comprehensive genome analysis into a single algorithm, identifying single nucleotide polymorphisms (SNPs), indels, CNVs, and structural variants (SV), with superior sensitivity and precision while reducing the time cost up to 72 fold. Comparative genomics studies typically limit their focus to SNVs, such as in previous comparisons of woolly mammoth and another comparison of eastern gorilla. We extended these analyses to identify SVs and indels. Our analysis found mammoth-specific variants suggesting adaptations to Arctic conditions, including variants associated with metabolism, immunity, circadian rhythms, and structural features. In gorilla populations, variants were identified that associate with physical features used to distinguish between the two subspecies. Within the gorilla subspecies was also found unique genetic evidence related to disease and abnormality, evidence of dwindling populations. Untested and ad hoc methods of mutation calling are often used in ancient DNA (aDNA) studies. While aDNA NGS analysis is highly susceptible to aDNA degradation, many studies utilize standard mutation calling algorithms, not taking into account unique aDNA challenges of excessive contamination, degradation, or environmental damage. We present ARIADNA, a novel approach based on machine learning techniques, using specific aDNA characteristics as features, to yield improved mutation calls. In our comparisons of variant callers across several ancient genomes, ARIADNA consistently detected higher-quality variants, while reducing the false positive rate compared to other approaches.

ETD_8636

Ph.D.

Includes bibliographical references

by Joseph Kenneth Kawash

doi:10.7282/T31Z47MJ

ETD doctoral

The author owns the copyright to this work.