OptiK: An Entropy-Driven Framework for Optimal k-mer Size Selection for Bacterial Genomics

K-mer-based approaches have become fundamental (Zielezinski et al., 2017) to modern computational genomics, underpinning tools for genome assembly, metagenomic classification, variant calling, and phylogenetic analysis. Despite their ubiquity, selecting an appropriate k-mer size (k) is often made arbitrarily or heuristically, with little consideration for the underlying signal quality relative to a given dataset. Here, I introduce OptiK, a novel alignment-free tool that evaluates the information richness of k-mer encodings across a range of k values to identify the optimal k for comparative analysis. OptiK operates by constructing k-mer frequency matrices from genome collections, reducing their dimensionality via truncated singular value decomposition (SVD), and evaluating clustering structure through unsupervised metrics including the Silhouette coefficient, Calinski-Harabasz index, and Davies-Bouldin index. We validate OptiK on a curated dataset of 1044 Helicobacter pylori genomes with well-characterized population structure. OptiK robustly identifies k = 8 as the optimal k-mer size, yielding latent structures in UMAP space that align with fineSTRUCTURE-defined subpopulations without relying on prior labels or reference alignments. These results demonstrate that OptiK provides a reproducible, alignment-free strategy for optimizing k-mer resolution in bacterial comparative genomics.