Classification of exon and intron regions obtained using digital signal processing techniques on the DNA genome sequencing with EfficientNetB7 architecture

Abstract

DNA is an important biomarker, containing enormous information for building the organism and maintaining its viability. DNA genome fragment with a symbolic sequence consisting of the letters A, T, G and C consists of protein-coding (exon) and non-coding (intron) parts. Identification of these regions plays an important role in different enlightening issues such as examining the development status of cancer, monitoring whether mutations occur in the relevant gene regions or regulating the growth and development of the organism. In this scope, it is aimed to distinguish the exon and intron regions correctly by computer-aided systems. In the first stage of the study, the most successful digital mapping technique on symbolic DNA sequences digitized with different numerical mapping techniques was decided by performance criteria. Then, the digitized DNA sequences using the mapping technique selected in the first part were expressed as spectrograms. Spectrograms, which are a visual representation of the frequency spectrum of a signal that changes over time, were labelled as exon and intron regions then were classified using the EfficientNetB7 model, a transfer learning architecture. At the end of the classification process, the success rate was obtained as %100.