Machine Learning Applied to Biological Sequence Comparison: an Alignment-free Approach

Biological sequence comparison is traditionally performed using algorithms that fall into the category of alignment approaches. These algorithms, however, have some limitations that can be overcome by alignment-free sequence comparison methods. Most of these alternative methods are based on word statistics or word comparison, after the biological sequences have been transformed into a set of subsequences of size k, called k-mer. For polypeptide sequences, feature extraction can also be done from the physicochemical qualities of the amino acids that compose them. Recently, many authors have used k-mer occurrences or k-mer frequencies, as well as physicochemical characteristics of amino acids, to train machine learning models. In this context, this work aimed to provide a comprehensive and initial guide to the use of the alignment-free approach, combined with machine learning algorithms, for the comparison of biological sequences. This article discusses the basic concepts and procedures of the alignment-free approach and machine learning, as well as provides a brief systematic review of recent literature to provide examples in the field. In addition, this guide is accompanied by 3 interactive online tutorials.