Controlling false-negative errors in microarray differential expression analysis: a PRIM approach.

TitleControlling false-negative errors in microarray differential expression analysis: a PRIM approach.
Publication TypeJournal Article
Year of Publication2003
AuthorsCole SW, Galic Z, Zack JA
Date Published2003 Sep 22
KeywordsAlgorithms, Artificial Intelligence, False Negative Reactions, Gene Expression Profiling, Oligonucleotide Array Sequence Analysis, Quality Control, Reproducibility of Results, Sensitivity and Specificity, Sequence Alignment, Sequence Analysis, DNA

MOTIVATION: Theoretical considerations suggest that current microarray screening algorithms may fail to detect many true differences in gene expression (Type II analytic errors). We assessed 'false negative' error rates in differential expression analyses by conventional linear statistical models (e.g. t-test), microarray-adapted variants (e.g. SAM, Cyber-T), and a novel strategy based on hold-out cross-validation. The latter approach employs the machine-learning algorithm Patient Rule Induction Method (PRIM) to infer minimum thresholds for reliable change in gene expression from Boolean conjunctions of fold-induction and raw fluorescence measurements.

RESULTS: Monte Carlo analyses based on four empirical data sets show that conventional statistical models and their microarray-adapted variants overlook more than 50% of genes showing significant up-regulation. Conjoint PRIM prediction rules recover approximately twice as many differentially expressed transcripts while maintaining strong control over false-positive (Type I) errors. As a result, experimental replication rates increase and total analytic error rates decline. RT-PCR studies confirm that gene inductions detected by PRIM but overlooked by other methods represent true changes in mRNA levels. PRIM-based conjoint inference rules thus represent an improved strategy for high-sensitivity screening of DNA microarrays.

AVAILABILITY: Freestanding JAVA application at

Alternate JournalBioinformatics
PubMed ID14512352
Grant ListAI 33259 / AI / NIAID NIH HHS / United States
AI 49135 / AI / NIAID NIH HHS / United States
AI 52737 / AI / NIAID NIH HHS / United States