IMPARO: inferring microbial interactions through parameter optimisation

• Published in: BMC cell biology Vol. 21; no. Suppl 1; p. 34
• Main Authors: Vidanaarachchi, Rajith, Shaw, Marnie, Tang, Sen-Lin, Halgamuge, Saman
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 19.08.2020; BioMed Central; BMC
• Subjects: Abundance; Accuracy; Algorithms; Bacteria - metabolism; Boolean; Data Accuracy; Feces - microbiology; Female; Gastrointestinal Microbiome; Healthy Volunteers; Humans; Inferring interactions; Intestinal microflora; Laboratories; Male; Mathematical models; Metagenomics; Microbial colonies; Microbial interaction network; Microbial Interactions - physiology; Microbiological research; Microbiomes; Microbiota (Symbiotic organisms); Models, Biological; Network dynamics; Observations; Statistical analysis; Statistical methods; Taiwan; Inferring interactions; Metagenomics; Microbial interaction network; Network dynamics
• ISSN: 2661-8850; 2661-8850; 1471-2121
• DOI: 10.1186/s12860-020-00269-y

Microbial Interaction Networks (MINs) provide important information for understanding bacterial communities. MINs can be inferred by examining microbial abundance profiles. Abundance profiles are often interpreted with the Lotka Volterra model in research. However existing research fails to consider a biologically meaningful underlying mathematical model for MINs or to address the possibility of multiple solutions. In this paper we present IMPARO, a method for inferring microbial interactions through parameter optimisation. We use biologically meaningful models for both the abundance profile, as well as the MIN. We show how multiple MINs could be inferred with similar reconstructed abundance profile accuracy, and argue that a unique solution is not always satisfactory. Using our method, we successfully inferred clear interactions in the gut microbiome which have been previously observed in in-vitro experiments. IMPARO was used to successfully infer microbial interactions in human microbiome samples as well as in a varied set of simulated data. The work also highlights the importance of considering multiple solutions for MINs.