Analysis of Microbiome High Dimensional Experimental Design Data using Generalized Linear Models and ANOVA Simultaneous Component Analysis

Fentaw Abegaz^1*Davar Abedini¹Lemeng Dong¹Johan Westerhuis¹Fred Van Eeuwijk²Harro Bouwmeester¹Age K. Smilde¹

• ¹University of Amsterdam, Amsterdam, Netherlands
• ²Wageningen University and Research, Wageningen, Netherlands

In microbiome studies, addressing the unique characteristics of sequence data such as compositionality, zero-inflation, overdispersion, high dimensionality, and non-normality is crucial for accurate analysis. In addition, it is worthwhile to integrate the experimental design elements into the microbiome data analysis for understanding how factors like treatment, time, and interactions affect microbial abundance. To achieve these objectives, we developed a new method that combines Generalized Linear Models (GLMs) with ANOVA Simultaneous Component Analysis (ASCA), which we coined GLM-ASCA. This method aims to improve microbiome analysis by providing a more comprehensive understanding of differential abundance patterns in response to experimental conditions. This is accomplished by modeling the unique characteristics of microbiome sequence data with GLMs and using ASCA to effectively separate the effects of different experimental factors on microbial abundance. GLM-ASCA is evaluated using simulated data and subsequently used on real data to analyze the effect of nitrogen deficiency on root microbiome recruitment in tomato. Simulation studies demonstrated the effectiveness of GLM-ASCA to analyze microbiome data in complex experimental designs, and the real data application revealed valuable insights into the dynamics of microbial communities under nitrogen starvation, as well as identifying beneficial bacterial species that promote the growth and health of tomato through nitrogen fixation.