A novel chemical property-based, alignment-free scalable feature extraction method for genomic data clustering

Abstract

In numerous fields of biological research, the precise clustering of genome sequences is of paramount importance. However, the inherent complexity and high dimensionality of genomic data produce substantial obstacles in achieving robust and efficient clustering results via traditional analysis, i.e., alignment-based approaches. The use of alignment-free approaches is one of the significant steps to perform the clustering efficiently. However, the majority of existing alignment-free approaches lack scalability, making it difficult to efficiently process the vast amount of genomic sequences. Moreover, the majority of approaches only extract the k-mer-based features and ignore the other significant features based on the classification of nucleotides according to their chemical properties. So, in order to handle these challenges, we present a novel scalable feature extraction method that efficiently handles large-scale genome data using the Apache Spark framework by distributing the tasks on various nodes and extracting the significantly important features based on the classification of nucleotides using their chemical properties in terms of entropy and the length of the sequence. The clustering of genome sequences is performed by taking extracted features using K-means, Fuzzy c-means, and Hierarchical agglomerative clustering. Our findings show that the proposed method improves generalization across many realistic plant genome and benchmark datasets and allows for the accurate clustering of formerly ambiguous cases. © 2025 Elsevier Ltd