Numeric Lyndon-based feature embedding of sequencing reads for machine learning approaches

Feature embedding methods have been proposed in the literature to represent sequences as numeric vectors be used some bioinformatics investigations, such family classification and protein structure prediction. Recent theoretical results showed that well-known Lyndon factorization preserves common factors overlapping strings [1]. Surprisingly, fingerprint of a sequencing read, which is sequence lengths consecutive variants effective capturing similarities, suggesting it basis for definition novel representations reads. We propose feature method Next-Generation Sequencing (NGS) data using notion fingerprint. provide experimental framework estimate behaviour fingerprints k-mers extracted from it, called k-fingers, possible embeddings As case study assess effectiveness embeddings, we use RNA-Seq reads order assign them most likely gene they originated fragments transcripts gene. an implementation tool lyn2vec, produces Lyndon-based