Skip to main content
. 2017 Sep 15;8(57):97025–97040. doi: 10.18632/oncotarget.20923

Figure 6. Examples of drugs having multi-gene markers with high test set recall and F-scores.

Figure 6

(Left) Mutated ERBB2 was the most significant single-gene marker for 17-AAG sensitivity (P=0.008) on the training set (the label “res” next to the gene indicate that ERBB2-mutant cell lines tend to be resistant to the drug). On the test set, this marker obtained a precision of 0.5 with practically no recall (RC=0.01). By contrast, the corresponding multi-gene marker has a similar level of precision with much higher recall (RC=0.54). Consequently, F-scores are also substantially higher for the multi-gene marker (F1=0.53 vs F1=0.03 of the single-gene marker). (Right) BCR_ABL translocation is the most significant single-gene marker for Methotrexate (P=0.0002). However, this gene fusion was not detected in any of the 234 test cell lines for this drug. For this drug, the multi-gene marker achieves a much higher test set precision and recall, which highlights that this approach is particularly useful when the single-gene marker is based on relatively rare gene mutations. In this case, F-scores are also substantially higher for the multi-gene marker (F1=0.63 vs F1=0 of the single-gene marker). Note the following abbreviations: “pred sens” (predicted sensitive) and “pred res” (predicted resistant).