ACFold & TFold tests

Contribution: Pinpointing differentlialy expressed proteins by only using statistical test can yield high false positive rates because such tests were not designed for massive hypotheses testing. Selecting only by fold changes can also lead to false conclusions. The ACFold & TFold combine average fold changes with statistics (AC-test and t-test respectively) and a theoretical false positive estimator to properly select the differentlialy expressed proteins. The software also provides options for handling labeled (e.g., SILAC) or unlabeled (e.g., spectral counting) data and can filter proteins that were detected within a minimum number of replicates.

The ACFold targets experiments with less than three replicates from each state or having assays acquired by different protocols while the TFold is recomended when replicate readings are obtained. Details of the method are given on PatternLabs' manuscript and furthe complemented in the GOEx manuscript.

In the TFold analysis in the figure below, each protein (represented as a dot) was mapped according to its log2(fold change) on the ordinate (y) axis and -log2(p-value)) on the abscissa (x) axis.  A total of 45 proteins (blue dots) were selected as differentially expressed because they satisfied both the t-test and the fold change cut-off. However, the estimates that only 40 of them are truly differentialy expressed. 22 proteins (orange dots) did not meet the fold change cutoff but were indicated as statistically differentially expressed, therefore deserving further analysis.  91 proteins (green dots) met the fold change cutoff, but the t- test indicated that this happened by chance. 1664 proteins (red dots) were pinpointed as not differentially expressed between classes because they failed both the t-test and the fold change cutoffs.