# A novel and efficient feature extraction algorithm using kmer-derived mutation signal

## TL;DR

This paper introduces a new genomic feature extraction algorithm using kmer-derived mutation signals to identify biologically significant genomic regions and improve classification and phylogenetic analysis.

## Contribution

A novel feature extraction algorithm using kmer-derived mutation signals to identify biologically significant genomic anchors and weighted interval entropy.

## Key findings

- The proposed algorithm outperforms position and frequency-based kmer methods in phylogenetic tree construction and machine learning classification.
- Anchor analysis in EBOV and HCV genomes reveals mutation types and directions, useful for diagnosing viral evolution.
- Weighted interval entropy effectively characterizes cumulative information content of genomic anchors.

## Abstract

Effective algorithms for extracting genomic features are crucial for downstream bioinformatics research. Although kmer-based descriptive statistical features (frequency and position) offer a unique perspective for genomic feature extraction, their biological significance warrants further investigation. Therefore, uncovering the biological significance of kmers in genomes remains a significant challenge in current kmer-based genomic feature mining. In this work, based on reverse, complementary and reverse-complementary of kmers to simulate genomic accumulation mutation behavior, we proposed a novel feature extraction algorithm. By examining the elastic length region following each kmer for the presence of any of these three scenarios, kmers of potential biological significance are identified and defined them as anchors of sequence. To characterize cumulative information content of anchor on genome, we defined a weighted interval entropy, based on the interval signal of the identified anchors. In performance evaluation, we compared our proposed algorithm against methods based on position and frequency of kmer. Our method demonstrates superior effectiveness in species phylogenetic trees and machine learning classification. Furthermore, we conducted an in-depth analysis of the types and position of anchor, in Ebola-Zaire virus (EBOV) and hepatitis C virus (HCV) genomes of different sampling times, indicating that anchor can be used to diagnose the type and direction of accumulation of HCV and Ebola mutations.