Abstract
1
Introduction
2
How bias affects abundance measurements
2.1
A model of MGS measurements
2.2
Relative abundance
2.3
Absolute abundance
2.3.1
Leveraging information about total-community abundance
2.3.2
Leveraging information about a reference species
3
How bias affects DA results
3.1
Fold change between a pair of samples
3.2
Regression analysis of many samples
3.3
Rank-based analyses
4
Case studies
4.1
Foliar fungi experiment
4.2
Vaginal microbiomes of pregnant women
4.3
Human gut microbiomes
4.4
Microbial growth in marine sediments
4.5
Summary and discussion
5
Solutions
5.1
Ratio-based relative DA analysis
5.2
Calibration using community controls
5.3
Absolute-abundance methods with more stable FEs
5.3.1
Use complementary MGS and total-abundance measurements
5.3.2
Normalize to a reference species
5.4
Bias sensitivity analysis
5.5
Bias-aware meta-analysis
6
Conclusion
Appendix
A
Linear regression
A.1
Simple linear regression
A.1.1
Review of simple linear regression
A.1.2
Measurement error in the response
A.1.3
Specific application to taxonomic bias
A.2
Gamma-Poisson regression
A.2.1
Background
A.2.2
Inferring LFCs in proportions with and without bias correction
B
Supplemental figures
Funding
Acknowledgments
References
Implications of taxonomic bias for microbial differential-abundance analysis
Acknowledgments
We thank Jen Nguyen and members of the Callahan Lab for valuable discussions.