GeneChip microarrays—signal intensities, RNA concentrations and probe sequences

Abstract

GeneChip microarrays consist of hundreds of thousands of oligonucleotide probes. Thetransformation of their signal intensities into RNA transcript concentrations requires theknowledge of the response function of the measuring device. We analysed the ‘apparatus’function of perfect match (PM) and mismatched (MM) oligonucleotide probes of GeneChipmicroarrays after changes of the target concentration using the results of a spiked-inexperiment. In agreement with previous studies we found that a competitivetwo-species Langmuir-adsorption model describes the probe intensities well. Each PMand MM probe is characterized by two hybridization constants which specify thepropensity of the probe to bind specific and non-specific transcripts. The affinity fornon-specific hybridization is on average equal for PM and MM. The purine–pyrimidineasymmetry of base pair interaction strengths, however, causes a characteristicPM–MM intensity difference, the sign of which depends on the middle base of theprobe. The affinity for specific hybridization of the PM exceeds that of the MM onaverage by nearly one order of magnitude because the central mismatched baseonly weakly contributes to the stability of the probe/target duplexes. For thefirst time we differentiate between the free energy parameters related to the 64possible middle-triples of DNA/RNA oligomer duplexes with a central Watson–Crickpairing and a central mismatched pairing. Both the PM and MM probes respondto the concentration of specific transcripts, which can be estimated from thePM and MM probe intensities using the Langmuir-model. The analysis of thePM–MM intensity difference provides at least no loss of accuracy and precision of theestimated concentration compared with the PM-only estimates which in turnoutperform the MM-only estimates. The results show that the processing of thePM–MM intensity difference requires the consideration of a background term dueto non-specific hybridization, which is, however, reduced by nearly one orderof magnitude when compared with the respective background of the PM andMM probes. The calculation of the sequence-specific affinity constants using apositional-dependent nearest-neighbour model opens up the possibility to estimate targetconcentrations beyond the training set of several hundred of spiked-in probes.