Fig. 5
From: Unearthing new genomic markers of drug response by improved measurement of discriminative power
Potential false-negative marker of the MANOVA test detected by the chi-squared test. (Left) The scatter plot for the drug-gene association (Dasatinib-CDKN2a.p14) with the largest -logPχ2 among those not significant according to the MANOVA test. Hence, mutated-CDKN2a.p14 is a potential marker of sensitivity to the marketed drug Dasatinib according to the chi-squared test, but not according to the MANOVA test. However, this marker has predictive value as it provides MCC = 0.13 on the test set. Therefore, the chi-squared test detected this potential false negative of the MANOVA test. (Right) Conversely, to assess the consistency of the MANOVA test, we searched for the drug-gene association with largest -logPMANOVA among those with a similar -logPχ2 to that of Dasatinib-CDKN2a.p14, which is SB590885-BRAF. Whereas the p-values for Dasatinib-CDKN2a.p14 and SB590885-BRAF differ in 27 orders of magnitude using the MANOVA test, the p-values for the same associations have similar p-values using the chi-squared test (Pχ2~ 10− 9). Thus, unlike the chi-squared test, the MANOVA test is unable to detect that both markers have similar discriminative power (SB590885-BRAF has a φ of 0.29 for 0.35 of Dasatinib-CDKN2a.p14). SB590885-BRAF is a true positive of both tests as its MCC on the test set is 0.27