Table 1S Detailed Information About 418 Animals Used in This Study

Supplementary Tables

Table 1S Detailed information about 418 animals used in this study.

Table 2S Summary of approaches used to compare data across platforms. Table A reflects the number of samples within each compound that have necrosis (+) and those that do not (-). Table B reflects the number of samples within each compound that have different RHI scores. Each data is denoted by platform immediately followed by an integer indicating the nature of the microarray data (1=intensity data and 2=ratio data) and the parenthesis with the number of the arrays; for example, AF1(318) denotes Affymetrix intensity data with 318 arrays.

A.

B.

Table 3SSummary of T-Index Scores and Prediction Accuracy Generated from Each of the 54 Permutations Tested for the Cross-Platform Transferability of Signature Genes. The analysis procedure for Nearest Centroid (NC), K-Nearest Neighbor (KNN) and Decision Forest (DF) was depicted in Supplementary Methods online

* AFX  AGL denotes the classifiers were developed on the Affymetrix platform and the signature genes then were transferred to the Agilent platform.
Table 4S Summary of T-Index Scores and Prediction Accuracy for the Cross-Platform Transferability of Signature GenesUsing Pathway Level Data and Performed Using K-Nearest Neighbor (KNN) Method.The analysis procedure was depicted in Supplementary Methodsonline.

* AFX  AGL denotes the classifiers were developed on the Affymetrix platform and the signature genes then were transferred to the Agilent platform.

Table 5SSummary of T-Index Scores and Prediction Accuracy Generated from Each of the 54 Permutations Tested for the Cross-Platform Predictivity of Classifiers.The analysis procedure was depicted in Supplementary Methods online

* AFX  AGL denotes the classifiers were developed on the Affymetrix platform and then predict the samples profiled by the Agilent platform.

** BatchMatch 2.0 (Systems Analytics Inc, was used to remove cross-platform batch effect for SVM and for both KNN and DF in analysis configuration 1. No cross-platform batch correction was applied for Analysis Configuration 2. Themean-centered batch correction was used in Analysis Configuration 3 for both DF and KNN.

Table 6SImpact of the Cross-Platform Batch Correction on the Prediction Accuracy Associated with Cross-Platform Predictivity of Classifiers. Results for SVM, DF and KNN before and after cross-platform batch correction.

* For analysis configuration 1, BatchMatch 2.0 (Systems Analytics Inc, was used for the cross-platform batch correction. For analysis configurations 2 and 3, the mean-centering was used for both KNN and DF while the BatchMatch algorithm was applied for SVM.

** AFX  AGL denotes the classifiers were developed on the Affymetrix platform and then predict the samples profiled by the Agilent platform that includes both Agilent training and test sets. AGL  AFX denotes a reverse approach was described above.