Effect of changing Z-score on Analysed Network Estimation.

The SynTReN simulated data for 100 samples was analysed using different Z-scores to select up- and down-regulated genes. Although the specificity increased at higher Z-scores the sensitivity was lower. Our strategy in looking for bio-markers is to accept relationships with lower significance at this stage but subsequently require that any useful pattern or clique is highly connected. In real situations, it is also important to require that the cliques are found in independent datasets. Our decision not to look at lower Z-scores than 0.4 is based on pragmatic biomarker requirements, where changes in expression have to be robust and indicate changes likely to be found by other methods.