Noise model estimation with application to gene expression

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene expression data analysis requires accurate noise estimation.
• Distinguishing biological variation from technical noise is crucial for reliable results.
• Existing methods may not adequately address processing errors in extracted data.

Purpose of the Study:

• To propose algorithms for estimating noise levels and detecting noise models in gene expression data.
• To develop methods for separating processing errors from biological noise in 1D gene expression profiles.
• To discriminate between additive and multiplicative noise models in both 1D and 2D gene expression data.

Main Methods:

• Development of algorithms for noise level estimation and noise model detection.
• Application of singular spectrum analysis (SSA) and its 2D extension for pattern extraction.
• Construction of an algorithm to separate processing errors from biological noise in 1D data.
• Testing algorithms on artificial data simulating real gene expression patterns.

Main Results:

• Algorithms successfully estimated noise levels and identified noise models in Drosophila embryo gene expression data.
• A method for separating processing errors from biological noise was effective for 1D data.
• Discrimination between additive and multiplicative noise models was achieved for both 1D and 2D data.
• Comparison of 1D and 2D methods showed distinct results for Krüppel and giant genes.

Conclusions:

• The proposed algorithms provide robust tools for noise analysis in gene expression data.
• Accurate noise estimation and model detection enhance the reliability of biological interpretations.
• The developed methods are applicable to both 1D and 2D gene expression datasets, offering flexibility in analysis.