Classification of coding and non-coding regions in eukaryotic gene sequences using an adaptive anti-notch filter...

Precise classification of coding regions from a gene sequence is extremely crucial, as they possess the most fundamental information regarding protein formation for numerous biological activities. The classification of these regions is immensely challenging, as these are generally distributed disparately and retain comparatively low volume density in eukaryotic genes. In this work, a data-independent, as well as window length-independent genomic signal processing technique is proposed to predict coding and non-coding regions by utilizing the period-3 characteristic of exons. At first, a finite impulse response-based anti-notch filter (ANF) is constructed to capture the single frequency ω₀=2π3 corresponding to period-3 characteristic. The filter is designed by adjusting its coefficients employing a hybrid Salp-Swarm-Whale optimization algorithm with a specially designed error function to replicate the ideal period-3 filter response. Subsequently, the symbolic gene sequence is transformed into binary data using Voss representation and then applied to the proposed ANF for capturing the period-3 frequency component. From the output signal of the ANF, high-frequency noise is eliminated using the low-pass Savitzky-Golay filter, and the power spectral density is estimated to evince period-3 peaks of coding regions. To substantiate the sequence length independence of the proposed method, classification performance is evaluated on short sequences of length ∼ 100 bp and long sequences of length ∼ 1000000 bp. Additionally, a new dataset comprising 191 sequences is created for evaluation to demonstrate the proposed method's dexterity on gene sequences from diverse categories of organism species. When evaluated using the benchmark gene sequence F56F11.4a, the method yielded an AUC value of 0.98 and the best accuracy value of 0.954 at a threshold value of 0.30. When evaluated on the newly created MOD191 dataset, comprising sequences from diverse organisms, the proposed method yielded an impressive AUC value of 0.91.