Accuracy prediction plots for Affymetrix first pass Sty HapMap samples. (a) A histogram of the BRLMM confidence measure is plotted for a sample chip with an average accuracy lower than 33% called by either BRLMM or CRLMM. (b) The graph shows a scatter plot of average accuracy of chips as called by BRLMM versus SNR. The y-axis is in the logit scale; the x-axis, the log scale.

Genotype regions. These plots display the space on which clusters are assigned genotypes. Colors represent HapMap gold standard calls; numbers represent the calls made by the array algorithms. For SNP SNP_A-1676170, (a) BRLMM genotype regions are shown in the Contrast Center Stretch (CCS) space. The x-axis is a contrast measure that captures the relative intensity difference of allele A with B. The y-axis is related to the average intensity from the A and B alleles. See the BRLMM white paper for more details [22]. For the same SNP, (b) CRLMM regions are plotted. The log ratios of allele A to allele B from intensities derived from probesets of the sense (y-axis) and antisense (x-axis) orientation are shown (that is, M₊ versus M_- plot). The ellipses represent the cluster regions obtained from the HapMap training set.

Comparison of SNP quality. CRLMM regions are plotted in log ratio space for (a) a high quality SNP (SNP_A-1750453) and (b) a low quality SNP (SNP_A-1709733). Hmz and Htz are abbreviations for homozygous and heterozygous, respectively.

Accuracy versus ADPs for CRLMM (orange) and BRLMM (purple). Drop rates between 0 and 10% are examined. (a) ADPs are plotted for 269 HapMap samples hybridized by Affymetrix on 100K chips (XBA and HIND). Only high-quality hybridization data, as defined by Affymetrix, are used in this plot. The gold standard is HapMap calls. (b) ADPs are plotted for first-pass data from 152 and 95 samples hybridized on the 500K chips (Nsp and Sty) from the Broad Institute and Affymetrix, respectively. Again, the gold standard is HapMap calls. (c) ADPs are plotted for the 32 and 16 high quality replicates hybridized on XBA and Nsp chips. The consensus of the 32 and 16 high quality replicate chips is considered the gold standard for each chip type. Separate gold standards are derived from each calling algorithm result. These data were generated from the Chakravarti Lab. (d) ADPs are plotted for the 30 high quality trios hybridized on XBA and Nsp chips. Accuracy for trios is defined as percent of SNP trios that are Mendelian consistent. The trios may not have any dropped calls. The data were generated from the Chakravarti Lab.

Robustness to bad quality chips. Accuracy is plotted against confidence thresholds for various datasets. In other words, the data in Figure 1 are plotted again except that the confidence measures used previously to achieve specified drop rates are now placed on the x-axis. Results of all HapMap datasets are shown from (a) BRLMM and (b) CRLMM. (c) Accuracy versus confidence plots (ACPs) are made for BRLMM (purple) and CRLMM (orange). The points are further stratified by call type according to the HapMap gold standard. The STY and NSP are the array types described in the text. Hmz and Htz are abbreviations for homozygous and heterozygous, respectively.

Comparison to Birdseed. (a) As Figure 1 but for 6.0 data and Birdseed instead of BRLMM. (b) As Figure 3 but for 6.0 data and Birdseed instead of BRLMM.

Log intensity ratios (allele A versus B), denoted with M, for all SNPs on one chip plotted against average log intensity S values. Both sense and antisense values are shown in all plots. A scatter-plot of these data would include 500,000 points and thus would be hard to interpret. We therefore show two-dimensional histograms with dark and light shades of blue indicating the existence of many and few points, respectively. (a) High quality array. (b) Low quality array.