Human Microbiome Dynamics: Causality Detection With Convergent Cross Mapping
Elisa B. Margolis, Antoinette R. Oot, David N. Fredricks · Open Forum Infectious Diseases · 2016
Background. Microbial communities that live in humans vary widely across hosts and often exhibit dramatic shifts and fluctuations over time. Attributing cause and effect to these changes is a challenge. Commonly used regression methods are prone to spurious correlations and are not optimal to infer interactions' directionality. Methods. Convergent Cross Mapping is an alternative method relying on the principle that cause must precede effect and can detect how consistently a shift in 1 microbe predicts another's response. We verified Convergent Cross Mapping performance with simulated microbiome data and value of its directionality for hypothesis testing with human microbiome studies. Results. In 2 examples we demonstrate that Convergent Cross Mapping can identify the drivers of a community function. We determined that among Lactobacillus species, only Lactobacillus crispatus and Lactobacillus iners constitutively lower vaginal pH. Using data from a study of Malawian infant microbiota we uncovered a particular Prevotella copri strain that potentially influences growth and recovery from malnutrition. Furthermore, Convergent Cross Mapping in longitudinal vaginal microbiome was able to clarify the directionality of interactions between 2 species that have been associated with treatment recurrence in bacterial vaginosis. Conclusion. Convergent Cross Mapping performed better with quantitative data (e.g. quantitative polymerase chain reaction) as compared to relative abundance data (e.g. microbiome studies based on next generation sequencing of broad-range 16S polymerase chain reaction or metagenomics). Still Convergent Cross Mapping proved a robust approach for inferring interactions and generating mechanistic hypotheses in a broad spectrum of longitudinal microbiome studies. Disclosures. All authors: No reported disclosures.