A-G.C.22: Graph-based Machine Learning approaches for predicting microbiome composition
Authors
Fatemeh Rajaei Nesheli
Queens university Belfast
chris Creevey
Queens university Belfast
Huiru(Jane) Zheng
Ulster university
John-Paul Wilkins
Queens university Belfast
Keywords
Microbiome, Graph-based Machine Learning, Taxonomy-free Classification, microbial community prediction, Functional Prediction
[sponser-meet-now-chat][/sponser-meet-now-chat]
Current microbiome research focuses on describing community composition and function within individual studies, but it remains limited in its ability to predict likely microbial communities and their functional potential given the presence of specific taxa or environmental changes. This limitation is particularly evident in studies based on 16S rRNA sequencing, which is widely used because of its cost-effectiveness but provides limited taxonomic resolution and only indirect functional insight.
Taxonomy-free frameworks such as Life Identification Numbers (LINs) provide a stable, genome-based representation of microbial diversity, where hierarchical relationships between lineages capture similarity at multiple resolutions, but their use remains underexplored in predictive microbiome modelling.
Recently the Creevey lab have developed a LIN-like taxonomy-free classification approach for 16S microbiome data for the entire Greengenes2 database. Here, we propose a computational framework that integrates this with machine learning for predictive microbiome analysis. First, we model microbial community structure by learning lineage-level co-occurrence patterns using graph-based approaches that leverage the hierarchical structure of LINs to predict coexisting lineages given a target lineage. Second, we extend this framework to functional inference by linking LINs to pangenome-derived gene content. Using graph-based propagation across related lineages, we estimate gene presence–absence profiles and aggregate these into community-level functional predictions mapped to metabolic pathways.
This framework provides a scalable and consistent approach for linking microbiome composition to function across heterogeneous environments and supports a shift from descriptive to predictive microbiome analysis using widely available 16S data.
Co-authors: Chris Creevey, Huiru Zheng, John-Paul Wilkins
Contact Attendee
Warning: Attempt to read property "user_email" on string in /home/1276969.cloudwaysapps.com/ydbgzhdjeq/public_html/wp-content/plugins/my-conference-now/functions.php on line 2826