Poster L15 - Creation and Validation of LIMÓN - Longitudinal Individual Microbial Omics Networks

Introduction: Microbial communities continually adapt to their surrounding environment. Such communities play pivotal roles in countless ecosystems from environmental to human health. Perturbations of these communities have been linked to a myriad of diseases, including mental health and neurological disorders, Chron’s disease and cancer. The function of each individual microorganism depends on their interactions with other members of their community. Furthermore, these multispecies ecosystems are dynamic making it essential to have robust methods to understand how alterations in their environment effect interactions among community members. Existing methods for identifying temporal microbial adaptations have been focused on abundance variations in individual taxa, leaving a crucial underexplored area of how microbial interactions evolve overtime and how dynamic microbial networks could be specific to each of the entities from which repeated measurements are taken from (e.g., different mice or individuals, distinct spatial locations). Doing so would require novel statistical approaches to handle the complicated nature of compositional repeated count data. To address this gap in knowledge, we have developed a pipeline, LIMÓN – Longitudinal Individual Microbial Omics Networks.

Materials and

Methods: Covariate corrected microbial values are extracted and undergo centered-log ratio normalization and network inference with Sparse inverse covariance Estimation for Ecological Association Inference (SPIEC-EASI) for each specified time point. Finally, individual networks from the overall aggregate network at each time point are estimated using Linear Interpolation to Obtain Network Estimates for Single Samples (LIONNESS).

Results, Conclusions, and Discussions: This provides sample specific networks that users can leverage to evaluate how individual network topology proprieties change over time (e.g., number of nodes, edges, communities, centrality characteristics). This novel platform includes methods to (1) handle dynamic temporal data, (2) account for sample covariates, and (3) identify sample-specific network characteristics from the overall aggregate network across time. In LIMÓN, longitudinal count data compositionality is removed using a zero-inflated negative binomial linear mixed model. Our approach allows users to perform sample covariate correction at this step to remove confounding effects, such as subject age or weight which is often necessary in human studies. Thus, LIMON provides a unique platform to reveal the heterogenous evolving relationships between microbial network properties and sample features of interest overtime.

Acknowledgements (Optional):