Effective gene prediction by high resolution frequency estimator based on least-norm solution technique.

Area of Science:

• Genomics
• Bioinformatics
• Digital Signal Processing

Background:

• Accurate identification of protein-coding regions in vast genomic sequences is crucial.
• Digital signal processing tools are increasingly applied to DNA feature extraction.
• Coding DNA segments exhibit a characteristic 3-base periodicity, unlike non-coding regions.

Purpose of the Study:

• To utilize the noise subspace concept for detecting hidden periodicities in DNA sequences.
• To develop and evaluate a least-norm estimator for enhanced gene prediction.
• To compare the proposed method with existing techniques like the sliding discrete Fourier transform (SDFT).

Main Methods:

• Application of linear algebra's subspace concept to spectrum estimation.
• Development of a least-norm estimator based on noise subspace.
• Comparative analysis using resolution, quality factor, sensitivity, specificity, miss rate, and wrong rate.

Main Results:

• The least-norm estimator demonstrates sharp period-3 peaks in coding regions.
• Background noise is effectively eliminated, clearly distinguishing coding from non-coding segments.
• The proposed method shows superior performance compared to the SDFT method in gene prediction across various organisms.

Conclusions:

• The noise subspace-based least-norm method offers an effective approach for accurate gene prediction.
• This technique significantly enhances the ability to identify protein-coding regions in DNA sequences.
• The proposed method provides a more accurate and robust alternative to existing gene prediction tools.