Improved gene prediction by principal component analysis based autoregressive Yule-Walker method.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Spectral analysis, particularly Fourier techniques, is widely used for gene prediction due to its simplicity.
• Model-based autoregressive (AR) spectral estimation offers higher resolution for small DNA segments, but accurate model order selection remains a challenge.

Purpose of the Study:

• To propose a novel technique combining Yule-Walker autoregressive (YW-AR) process with principal component analysis (PCA) for improved gene prediction.
• To enhance the detection of protein-coding regions in DNA by analyzing spectral peaks at the 1/3 frequency component.
• To overcome the critical issue of optimal model order selection in AR spectral estimation.

Main Methods:

• A hybrid approach integrating YW-AR spectral estimation with PCA for dimensionality reduction.
• Utilizing the eigenvalue ratio to establish a threshold for separating signal and noise subspaces, facilitating data reduction.
• Employing PCA for noise removal before power spectral density (PSD) estimation, thus mitigating the need for precise model order selection.

Main Results:

• The proposed YW-AR and PCA combined method demonstrates superior performance in detecting protein-coding regions compared to traditional Fast Fourier Transform (FFT) and Wavelet Packet Transform (WPT) combined autoregressive methods.
• Receiver Operating Characteristics (ROC) and Discrimination Measure (DM) analyses confirm the effectiveness and superiority of the proposed technique.
• Noise reduction by PCA prior to PSD estimation makes optimal model order selection less critical.

Conclusions:

• The integration of YW-AR and PCA presents a robust and effective method for gene prediction, particularly for identifying protein-coding regions.
• This approach offers improved resolution and noise reduction, overcoming limitations of existing spectral analysis techniques in genomics.
• The method provides a more reliable alternative for DNA signal analysis in bioinformatics applications.