Microbial Interaction Prediction in Health and Disease

The human microbiome comprises diverse microbial communities whose interactions with each other and with the host play a pivotal role in maintaining health and in the emergence of disease. Computational prediction of these interactions addresses the complexity and scale of experimental screening by integrating genomic, phenotypic and network data. By modelling the topology of microbe–microbe and microbe–host networks, machine-learning frameworks can uncover candidate associations that inform diagnostics, probiotic design and targeted therapies. Advances in similarity measures, graph-based architectures and neural embedding have driven recent gains in predictive accuracy, while cross-disciplinary approaches are beginning to translate in silico predictions into clinical and environmental applications. This field is global in scope, impacting fields from infectious disease surveillance to metabolic disorder management, and it continues to evolve through the synthesis of bioinformatics, systems biology and artificial intelligence.

Research from Nature Portfolio

A foundational semi-supervised framework introduced Laplacian regularised least squares to predict human microbe–disease associations by combining Gaussian interaction profile kernel similarity with network smoothness constraints. This approach achieved robust performance across global and local cross-validation schemes and reliably prioritised microbes linked to asthma, colorectal carcinoma and chronic obstructive pulmonary disease. The model’s validation against manual literature curation demonstrated its potential to guide experimental efforts and to deepen understanding of microbe involvement in disease mechanisms.

Microbial Interaction Prediction in Health and Disease publication trend

The graph below shows the total number of articles in microbial interaction prediction in health and disease across all publications each year (not limited to Nature Index journals).

Technical terms

Heterogeneous network: A graph comprising nodes and edges of multiple types, representing diverse biological entities and their relationships.

Graph neural network: A machine-learning architecture designed to learn representations from graph-structured data by propagating and aggregating information across nodes and edges.

Laplacian regularisation: A technique that imposes smoothness on predictions across a network by penalising large differences between connected nodes.

Gaussian interaction profile kernel: A similarity metric that quantifies relatedness based on known interaction patterns, using a Gaussian function to weight profile proximity.

Cross validation: A statistical method for evaluating predictive models by partitioning data into training and test subsets to assess generalisability.

References

1. LRLSHMDA: Laplacian Regularized Least Squares for Human Microbe–Disease Association prediction. Scientific Reports (2017).
2. Multi-scale topology and position feature learning and relationship-aware graph reasoning for prediction of drug-related microbes. Bioinformatics (2024).
3. Identifying Microbe-Disease Association Based on a Novel Back-Propagation Neural Network Model. IEEE Transactions on Computational Biology and Bioinformatics (2021).