Skip to main content

Table 3 Type I error control and power on simulated data based on Saccharibacteria’s normalized abundance

From: Powerful and robust non-parametric association testing for microbiome data via a zero-inflated quantile approach (ZINQ)

Sample size = 600      
  Type I error Power
  Null Setting 1 Setting 2 Setting 3
α-level 0.05 0.01 0.05 0.05 0.05
Rarefaction      
Linear regression 0.0247 0.0032 0.0602 + 0.0424 + 0.0326 +
ZIP 0.8238 0.7766 0.8056 0.7642 0.7384
ZINB 0.4241 0.2916 0.3515 0.3304 0.3150
ZINQ-MinP 0.0471 0.0089 0.9243 0.4867 0.0819
ZINQ-Cauchy 0.0506 0.0100 0.9166 0.5428 0.0954
TSS      
Linear regression 0.0279 0.0030 0.0372 + 0.0338 + 0.0320 +
ZIB 0.0053 0.0009 0.0649 + 0.0190 + 0.0067 +
Tobit 0.0522 0.0137 0.0837 0.0703 0.0635
ZIlogN 0.9997 0.9997 0.9987 0.9978 0.9983
ZIG 0.0495 0.0073 0.1094 0.0675 0.0498
ZINQ-MinP 0.0428 0.0083 0.6626 0.2480 0.0596
ZINQ-Cauchy 0.0497 0.0099 0.6800 0.2818 0.0700
CSS      
Linear regression 0.0500 0.0107 0.2021 0.1034 0.0541
Tobit 0.0498 0.0111 0.1677 0.0929 0.0533
ZIlogN 0.0446 0.0071 0.4933 0.2063 0.0621
ZIG 0.0443 0.0076 0.5563 0.2264 0.0643
ZINQ-MinP 0.0456 0.0085 0.8442 0.3766 0.0720
ZINQ-Cauchy 0.0497 0.0099 0.8327 0.3897 0.0773
  1. Setting 1: 100% from HBP edf for HBP samples;
  2. Setting 2: 80% from HBP edf and 20% from non-HBP edf for HBP samples;
  3. Setting 3: 60% from HBP edf and 40% from non-HBP edf for HBP samples.
  4. : power of a method that inflates type I error
  5. +: power of a method that deflates type I error
  6. Results by the various methods on 10000 simulated datasets by generating samples from the edf of Saccharibacteria’s normalized abundance, including type I error control and power under different settings with significance cutoffs 0.05 and 0.01