Wednesday, April 15 2020

9:00am - 11:30am

9:00am - 11:30am

Masters Presentation

Title

Exploring Gao et al. as a method for finding the effective number of independent tests in metabolomic data

Exploring Gao et al. as a method for finding the effective number of independent tests in metabolomic data

Abstract

When studying metabolomics data, thousands of tests can be run on the same dataset, which creates the multiple testing problem. Two popular methods that adjust for multiple testing are Bonferroni’s correction and False Discovery Rate (FDR). The Bonferroni method can be quite conservative if there is a correlation in a dataset, which could lead to a higher rate of false negatives. Instead, an effective number of independent tests can be estimated using a method by Gao et al and applied with Bonferroni’s correction to help make the correction less conservative. In this paper, I first simulated data to evaluate how well Gao’s method performs at finding the effective number of independent tests for datasets with different pairwise correlation structures. The simulation results suggest that if a dataset has more predictors than observations then the effective number of independent tests will be underestimated. Second, I applied Gao’s method to a case study with 4817 metabolites and 40 observations estimating 39 effective independent tests. This is further evidence that Gao’s method underestimates the effective number of independent tests when the number of predictors is greater than the number of observations in a dataset. Overall, I uncover that Gao’s method performs well at finding the effective number of independent tests when the number of observations is much greater than the number of predictors being studied. But the method fails when the number of observations is not considerably larger than the number of predictors.

When studying metabolomics data, thousands of tests can be run on the same dataset, which creates the multiple testing problem. Two popular methods that adjust for multiple testing are Bonferroni’s correction and False Discovery Rate (FDR). The Bonferroni method can be quite conservative if there is a correlation in a dataset, which could lead to a higher rate of false negatives. Instead, an effective number of independent tests can be estimated using a method by Gao et al and applied with Bonferroni’s correction to help make the correction less conservative. In this paper, I first simulated data to evaluate how well Gao’s method performs at finding the effective number of independent tests for datasets with different pairwise correlation structures. The simulation results suggest that if a dataset has more predictors than observations then the effective number of independent tests will be underestimated. Second, I applied Gao’s method to a case study with 4817 metabolites and 40 observations estimating 39 effective independent tests. This is further evidence that Gao’s method underestimates the effective number of independent tests when the number of predictors is greater than the number of observations in a dataset. Overall, I uncover that Gao’s method performs well at finding the effective number of independent tests when the number of observations is much greater than the number of predictors being studied. But the method fails when the number of observations is not considerably larger than the number of predictors.

Speaker: | Valentinas Sungaila |

Affiliation: | |

Location: | See Zoom link from email |

Done