[Paper Review] Removing System Noise from Comparative Genomic Hybridization Data by Self-Self Analysis

Yoon-ha Lee, Michael Ronemus+9 more|arXiv (Cornell University)|May 4, 2011

Genomic variations and chromosomal abnormalities25 references3 citations

TL;DR

This paper proposes a system normalization method to reduce technical noise in comparative genomic hybridization (aCGH) data by analyzing self-self hybridizations—where a DNA sample is hybridized to itself—using singular value decomposition (SVD). The approach identifies and corrects for system noise, improving signal-to-noise ratios by 7.0% on average, with 90% of hybridizations showing enhanced signal quality and reduced median average deviation (MAD).

ABSTRACT

Genomic copy number variation (CNV) is a large source of variation between organisms, and its consequences include phenotypic differences and genetic disorders. CNVs are commonly detected by hybridizing genomic DNA to microarrays of nucleic acid probes. System noise caused by operational and probe performance variability complicates the interpretation of these data. To minimize the distortion of genetic signal by system noise, we have explored the latter in an archive of hybridizations in which no genetic signal is expected. This archive is obtained by comparative genomic hybridization (CGH) of a sample in one channel to the same sample in the other channel, or 'self-self' data. These self-self hybridizations trap a variety of system noise inherent in sample-reference (test) data. Through singular value decomposition (SVD) of self-self data, we have determined the principal components of system noise. Assuming simple linear models of noise generation, the linear correction of test data with self-self data -or 'system normalization'- reduces local and long-range correlations and improves signal-to-noise metrics, yet does not introduce detectable spurious signal. Using this method, 90% of hybridizations displayed improved signal-to-noise ratios with an average increase of 7.0%, due mainly to a reduced median average deviation (MAD). In addition, we have found that principal component loadings correlate with specific probe variables including array coordinates, base composition, and proximity to the 5' ends of genes. The correlation of the principal component loadings with the test data depends on operational variables, such as the temporal order of processing and the localization of individual samples within 96-well plates.

Motivation & Objective

• To identify and characterize system noise in comparative genomic hybridization (aCGH) data arising from technical variability in sample processing and probe performance.
• To develop a noise correction method that preserves true biological signals while minimizing distortion from technical artifacts.
• To establish a systematic normalization framework using self-self hybridizations as a reference to model and correct for system noise in test data.
• To investigate the relationship between noise components and probe-specific variables such as array coordinates, base composition, and gene location.
• To validate that the correction method does not introduce spurious signals while improving signal-to-noise metrics in real aCGH experiments.

Proposed method

• Collect self-self aCGH data by hybridizing the same DNA sample to both channels of a microarray, creating a control dataset with no biological signal but containing all system noise.
• Apply singular value decomposition (SVD) to the self-self data to extract the principal components representing dominant noise patterns.
• Model system noise as a linear combination of these principal components, assuming noise generation follows a simple linear process.
• Use the identified noise components to linearly correct test aCGH data through system normalization, adjusting for probe-specific and operational noise sources.
• Correlate principal component loadings with probe-level variables (e.g., array coordinates, GC content, 5' proximity to genes) to understand noise origins.
• Assess correction performance using signal-to-noise metrics, including median average deviation (MAD), before and after normalization.

Experimental results

Research questions

• RQ1 To what extent can system noise in aCGH data be modeled and removed using self-self hybridization controls?
• RQ2 How do probe-specific features such as GC content and genomic location correlate with dominant noise components?
• RQ3 Does the proposed system normalization method improve signal-to-noise ratios without introducing false positive copy number changes?
• RQ4 How does the temporal order of sample processing and plate localization affect noise patterns in aCGH data?
• RQ5 Can the principal components of noise from self-self data be reliably used to correct real test data across multiple hybridizations?

Key findings

• System normalization using self-self data improved the signal-to-noise ratio in 90% of hybridizations, with an average increase of 7.0%.
• The primary improvement came from a reduction in median average deviation (MAD), indicating lower variability in corrected data.
• Principal component loadings of noise showed significant correlation with probe-specific variables, including array coordinates, GC content, and proximity to 5' ends of genes.
• Noise patterns were influenced by operational variables such as the temporal order of processing and sample localization within 96-well plates.
• The correction method did not introduce detectable spurious signals, confirming its fidelity in preserving true biological variation.
• SVD of self-self data effectively captured the dominant sources of system noise, enabling accurate linear correction of test data.