Nucleotide distribution variance-based dynamic representation scheme for novel gene prediction

Abstract

Predicting novel genes is an important topic in bioinformatics. De novo protein coding region prediction techniques are more powerful than homology-based techniques in the analysis of novel DNA sequences for annotating protein regions. In this article, a new gene finding technique is proposed to predict protein coding regions in DNA sequences. The technique is based on the spectral analysis of DNA sequences and calculating the period-3 spectrum. An analysis of the effectiveness of the nucleotides is performed to study the variance in the strength of the nucleotides in the period-3 spectrum in protein coding and non-coding regions. The proposed technique uses a dynamic representation scheme to map DNA sequences into a numerical form. The dynamic representation scheme provides better differentiation between coding and non-coding regions as it enhances the participation of nucleotides that are effective in the period-3 spectrum. The technique also uses post-processing to detect the period-3 spectrum peaks instead of thresholding. The proposed technique is compared with other spectrum-based techniques by plotting the receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). The results show that the proposed technique outperforms other spectrum-based techniques. In addition, we analyze the false positive peaks that result in from the prediction by scanning the availability of stop-codon-like triplet combinations in the three possible reading frames. This analysis provides an insight for future work to improve the performance of the spectrum-based gene finding techniques.