Predicting Vemurafenib Resistance in Melanoma

In our previous study, we identified 18 human proteins differentially expressed between vemurafenib-resistant A375 cells and compared to parental using a Human XL Oncology Array (Cat# ARY026; R&D Systems, Minneapolis, MN, USA) [22 (link)]. A total of 262,143 permutations/combinations between the 18 genes were tested to predict the resistance stage in GSE99898 samples, and validated in the other datasets and the “FiveDatabase”. Classifier markers were estimated after a receiver operative curve (ROC) analysis performed with the OptimalCutpoint R package (version 1.1-5). The area under the curve (AUC), positive predictive value (PPV), negative predictive value (NPV), sensitivity, and specificity were used as classifier metrics. The combination of genes with the highest AUC was selected.
A score, which was named the 13-gene risk score, was calculated according to the following equation: $\sum x_i\ast {\beta }_i$ , with $x_i$ = expression value of DEG $i$ , and ${\beta }_i$ = regression coefficient of gene $i$ from a linear discriminant analysis based on pre-treatment (sensitive) or progression (resistant) stage.
A fold change (FC) of the 13-risk score between PROG and PRE samples was calculated for each sample and stratified as “Down” if FC < 0, “Slightly up” if 0 < FC < z, where z = 0.25 * max (FC), and “Up” if FC > z.