Variation between channels in the distribution of signal intensities for NCPs. Cumulative distribution functions of signal intensities for NCPs for different channels are shown.

Linear dependence of signal intensity on GC content for negative control probes. Average log₂ signal intensity of NCPs with indicated GC content, for six channels.

Signal intensity distributions of probes measuring annotated A. gambiae exons (EPs), non-exon regions (NEPs) and negative controls (NCPs). Cumulative distribution functions of signal intensities for these probe populations are shown for a representative channel.

ROC curves showing true positive rate versus false positive rate relative to transcripts annotated in the UCSC database. The '+' symbol corresponds to the transfrags as defined by Cheng et al. [3]. Lines correspond to our algorithm as applied with/without neighborhood smoothing and with/without minrun/maxgap post-processing.

Combining information from replicate experiments. (a) Distribution of Fisher p values for NCPs, obtained by taking the product of channel-specific p values and comparing it to the product of random numbers drawn from a uniform distribution (see [28]). The solid line corresponds to probe signal intensities that were first corrected for probe sequence bias using the Full Position-specific model (see Materials and methods); the dashed line corresponds to uncorrected intensities. (b) Distribution of the negative log of products of single-channel p values for different probe populations.

Schematic overview of the TranscriptionDetector data processing pipeline. First, a model accounting for the effect of probe sequence on non-specific binding is fit to the (log-transformed) NCP signal intensities ('step 1') and used to correct the intensity for all probes ('step 2'); a separate model is fit for each channel. For each probe, we then derive a p value reflecting the likelihood that its signal intensity belongs to the background distribution represented by the NCPs ('step 3'); these p values are calculated separately for each channel, and each channel-specific p value is treated as the outcome of an independent experiment. A multi-channel statistic equal to the product of p values across all channels is computed for each probe ('step 4'). In analogy with step 3, the distribution of this statistic for the NCPs only is then used to assign a MCPV to the other probes ('step 5'). To control for multiple hypothesis testing, a FDR procedure is used, and each probed locus is designated as transcribed or not transcribed ('step 6').