ISCAZIM: Integrated statistical correlation analysis for zero-inflated microbiome data

Microbiome-metabolome association analysis is critical to reveal the key pairs of gut microbiota and metabolites for discovery of the microbial biomarkers in chronic diseases. However, the characteristics of microbiome data, such as zero inflation, over dispersion, may impair the confidence of association analysis between microbiome and metabolome data. The objectives of this study are to evaluate the strengths and weaknesses of existing statistical methods and to develop a computational framework tailored to the unique characteristics of microbiome data. We designed a computational framework called Integrated Statistical Correlation Analysis for Zero-Inflated Microbiome data (ISCAZIM) that takes account of complicated microbiome data characteristics, including zero inflation rates (ZIRs), dispersion and correlation patterns. ISCAZIM first benchmarked prevalent statistical correlation methods, Pearson, Spearman, zero inflated negative binomial (ZINB) model, mutual information and Maximal Information Coefficient. ISCAZIM then classifies the correlation pattern to linear or non-linear and applies the correlation method according to the ZIRs status. Applying to multiple real-world microbiome-metabolomics data, ISCAZIM is overall more accurate than using a single method with more truly significant association pairs included. Therefore, ISCAZIM will significantly facilitate the association analysis using zero-inflated microbiome data for multi-omics integration.