Estimation and correction of non-specific binding in a large-scale spike-in experiment

Area of Science:

• Bioinformatics
• Genomics
• Statistical Modeling

Background:

• Non-specific binding (NSB) significantly impacts microarray data, affecting 30-50% of probes.
• Accurate NSB estimation is vital for reliable normalization and statistical analysis in Affymetrix whole genome chip experiments.

Purpose of the Study:

• To benchmark various models for estimating NSB using a large-scale spike-in dataset.
• To evaluate the influence of probe sequence on NSB signals.

Main Methods:

• Benchmarking of MAS5 perfect match-mismatch (PM-MM), Naef, and Zhang sequence-based models.
• Utilizing a large-scale spike-in microarray dataset for model evaluation.
• Analysis of intensity-dependent performance of different methods.

Main Results:

• MAS5 PM-MM is a poor estimator of NSB; Naef and Zhang models provide reasonable estimates.
• GC-NSB generally outperforms other methods for differential expression detection.
• Optimal methods for probeset expression vary by intensity: GC-NSB (low), MAS5 PM-MM (medium), MAS5 PM-MM and PDNN (high).

Conclusions:

• Combining MAS5 PM-MM, GC-NSB, and PDNN methods improves differential expression detection and false discovery rate estimation.
• Eliminating empty probesets before normalization further enhances differential expression analysis.
• Further research with known transcript concentrations is needed to develop better specific binding models for Affymetrix GeneChip technology.