A Comparative Analysis between K-mer Database Methods and Traditional Approaches for Genotyping

The rapid growth of next-generation sequencing (NGS) throughput has increasingly strained conventional alignment-based genotyping pipelines, which rely on computationally expensive read mapping and variant calling steps. Alignment-free, k-mer-based methods have emerged as a promising alternative, bypassing full read alignment to reduce runtime and resource demands. This thesis evaluates GenoLight, a mapping-free Single Nucleotide Polymorphism (SNP) genotyping tool that replaces conventional k-mer dictionaries with a compressed reassembly index derived from k-mer set assembly, addressing the prohibitive memory requirements of earlier alignment-free tools such as LAVA and VarGeno. GenoLight was benchmarked against the standard alignment-based pipeline using the NA12878 genome and its Genome in a Bottle high-confidence variant set as ground truth. Two GenoLight configurations were tested, one indexed against the dbSNP catalogue (GL-dbSNP) and one against the smaller Affymetrix SNP Array 6.0 catalogue (GL-Affy), across three sequencing depths (high, mid, and low coverage). Performance was assessed in terms of CPU runtime, peak memory usage, and genotyping accuracy (precision, sensitivity, and F-measure), and each GenoLight configuration was additionally compared against the standard pipeline's own call set to quantify concordance. Results show that GenoLight achieves substantial and reproducible computational gains, with speedups ranging from 3.22× to 6.04× over the standard pipeline and memory footprints as low as 5.8 GB for GL-Affy, below the standard pipeline's own usage. However, these gains come at the cost of reduced sensitivity, particularly for GL-Affy, whose sensitivity is fundamentally constrained by the scope of its underlying catalogue rather than by sequencing depth. GL-dbSNP offers an intermediate trade-off, retaining a majority of gold-standard variants at high precision but with a growing false-positive rate at lower coverage. These findings indicate that no single method dominates universally: the standard pipeline remains preferable for comprehensive variant discovery at high coverage, while GenoLight is best suited as a targeted, resource-efficient genotyping accelerator for well-catalogued variant sets.