On the utility of pooling biological samples in microarray experiments

Kendziorski et al. 10.1073/pnas.0500607102.

Supporting Information

Files in this Data Supplement:

Supporting Text
Supporting Figure 7
Supporting Figure 8
Supporting Figure 9
Supporting Figure 10
Supporting Figure 11
Supporting Figure 12
Supporting Figure 13
Supporting Figure 14

Supporting Figure 7

Fig. 7. RMA fits a linear model to the log probe intensities for each probe set. The linear model includes a sample effect (the parameter of interest), a probe effect, and an error term. Shown here are standard error estimates of the sample effects derived from the linear model used in RMA (described in Preprocessing and Normalization). The standard errors are normalized so that each probe set has the same median standard error across arrays.

Supporting Figure 8

Fig. 8. Hierarchical clustering dendrogram of pools of two, pools of three, and averages across individuals for MAS 5.0 signal values. Common colors are given to pools and their corresponding averages. With the MAS 5.0 signal values, we see little correspondence among averages and pools. Compare these findings to the robust multiarray analysis data in Figs. 9 and 10, where pools and averages are clustered closely together. Euclidean distance was used to determine distances; the complete linkage method was used to construct the dendrogram.

Supporting Figure 9

Fig. 9. Hierarchical clustering dendrogram of pools of two and averages across individuals for robust multiarray analysis (RMA) data. Common colors are given to pools and their corresponding averages. Nearly perfect concordance is observed between the pools and their averages. A similar dendrogram can be shown for pools of three. Euclidean distance was used to determine distances; the complete linkage method was used to construct the dendrogram.

Supporting Figure 10

Fig. 10. Hierarchical clustering dendrogram of pools of two, pools of three, and averages across individuals for RMA data. Common colors are given to pools and their corresponding averages. There is not perfect, but still good, concordance among pools and their corresponding averages. Euclidean distance was used to determine distances; the complete linkage method was used to construct the dendrogram.

Supporting Figure 11

Fig. 11. Density estimates of the cumulative distribution functions shown in Fig. 2.

Supporting Figure 12

Fig. 12. As detailed in Preprocessing and Normalization, Fig. 6 was produced from data files (individuals and pools of 2, pools of 3, and pools of 12) that were processed together by using RMA. Shown here is Fig. 6 reproduced from data where RMA was applied separately within the individuals and pools of 2, pools of 3, and pools of 12. The results are virtually identical.

Supporting Figure 13

Fig. 13. M vs. A plots of pools of two (P) and the corresponding mathematical averages (V). Here, M = log₂(P/V ) and A = 0.5×log₂ (P ×V ). (Left and Center) Averages compared to the corresponding pools. (Right) Averages plotted against a noncorresponding pool.

Supporting Figure 14

Fig. 14. Comparison of 12 on 6 vs. 7 on 7 designs (see Comparison of Designs for more details on the construction of this figure).