Gene Prediction Based On All-phase Digital Signal Processing

DNA is the carrier of genetic material. It is a combination with four bases in sequence. In a DNAsequence, only some special regions could be encoded into protein. These regions are called as exons in agene. And those other regions that could not be encoded into protein are called as introns. The purpose ofgene identification algorithm is to detect all the exons from an unknown DNA sequences.In this thesis, the method of digital signal processing is used to do the gene prediction. The first step ofprocessing the gene signal is to do numerical mapping. It could transfer the alphabetic sequences tonumerical sequences. Then the method based on signal processing could be done to analyze the frequencydomain characteristics of the gene sequence. Because of the existence of the3-periodicity in these exonregions, which are always caused by codon bias in the coding region, DSP could be used as a method todistinguish exon and intron.To the issue of numerical mapping of DNA sequence, some common numerical mappings of DNAsequence are analyzed. The three most commonly used numerical mapping with definite physical meaningis introduced. The fast algorithm in computing the power spectrum and signal-to-noise ratio is deduced. Atthe same time the relationship of them and the closed-form expression of matrix are given.To the issue of gene prediction, the traditional spectral analysis would have spectrum leakage when thedata is truncated. That would strongly affect the accuracy of gene identification. To avoid the truncationeffect, and reduce the background noise, all-phase digital signal processing is combined in the geneprediction algorithm. The first step of this technology is the all-phase data preprocessing. This measurecould keep the continuity of the two side of the waveform and largely reduce the truncation effect. In thisthesis, a gene prediction algorithm based on all-phase FFT analysis is given. It could extract the3-periodicity signal in the gene and enhance the accuracy of gene identification. At the same time themultirate signal processing model is also combined. It could reduce the computation complexity andprovide a scheme for pipeline operation.In order to make further efforts to extract the3-periodicity signal in the DNA sequences and enhance theaccuracy of gene identification, the all-phase FIR model is introduced. In this thesis, an all-phase narrowband filter for gene prediction is designed. Because all-phase FIR could strictly keep the linear phase andinvariance phase, the phase rotation measure is combined for more accurate prediction results. Quantifiedevaluating indicator is introduced to compare the results from different gene prediction algorithms.Experimental results show that this algorithm has a high level of prediction accuracy at the nucleotide level.