AI for microbiome modelling

Synthetic Microbial Community
Synthetic Microbial Community

Plant diseases continue to threaten global crop yields and agricultural sustainability. Traditional reliance on chemical pesticides not only poses environmental and human health risks but also drives the evolution of resistant pathogens. A promising alternative lies in harnessing the plant-beneficial microbiome to enhance disease resistance, soil fertility, and overall productivity in a sustainable manner.

One of our lab’s key research directions focuses on the root microbiome, a structured microbial community that assembles in a deterministic way under consistent environmental and host conditions. Compared to bulk soil, root-associated microbiomes are less diverse—suggesting strong host-mediated selection and microbial competition. Microbial taxa with traits such as motility, chemotaxis, biofilm formation, and specific metabolic capabilities—often associated with families like Rhizobiaceae—have been linked to successful root colonization (root competence).

However, current methods to study root competence and microbiome assembly face important limitations. Genome-based annotations often rely on known gene functions, which restricts discovery. Functional annotations like Gene Ontology (GO) lack specificity, while pathway-level annotations such as KEGG Orthology (KO) only cover a fraction of bacterial genomes. More comprehensive systems like OrthoFinder orthologous groups (OGs) yield extremely high-dimensional datasets, complicating predictive modeling.

Our lab explores artificial intelligence (AI) methods to overcome these challenges. We develop machine learning models that predict microbial traits such as root competence and environmental specificity directly from genomic and metagenomic data. We investigate approaches that can handle high-dimensional input spaces, generalize across diverse microbial genomes, and integrate weakly annotated datasets.

Root microbiome assembly is not only driven by individual microbial traits but also by the community context: microbe-microbe interactions, host genotype, and environmental variables all play critical roles. Traditional modeling efforts have primarily focused on metabolic interactions within small synthetic communities, overlooking non-metabolic interactions and broader ecological dynamics.

To address this, we are developing multi-modal AI frameworks that combine omics data, host plant characteristics, environmental parameters, and microbial interaction networks. These integrative models aim to provide better predictive accuracy and broader applicability to real-world agricultural systems.

Our goal is to unlock the full potential of the plant microbiome by developing scalable AI tools that support the design of resilient, disease-suppressive crop systems.

Researchers

Publications

  • Dai, R., Zhang, J., Liu, F., Xu, H., Qian, J.-M., Cheskis, S., Liu, W., Wang, B., Zhu, H., & Pronk, L. J. (2025). Crop root bacterial and viral genomes reveal unexplored species and microbiome patterns. Cell.
  • de Jonge, R., & Selten, G. (2025). Bacillus: Driver of functional states in synthetic plant root bacterial communities. BioRxiv, 2025–03.
  • Selten, G., Gomez Repolles, A., Lamouche, F., Radutoiu, S., & de Jonge, R. (2025). SyFi: Generating and using sequence fingerprints to distinguish SynCom isolates. BioRxiv, 2025–02.
  • Selten, G., Lamouche, F., Gomez-Repolles, A., Blahovska, Z., Kelly, S., de Jonge, R., & Radutoiu, S. (2024). Functional capacities drive recruitment of bacteria into plant root microbiota. BioRxiv, 2024–08.
  • Song, Y., Atza, E., Sánchez-Gil, J. J., Akkermans, D., de Jonge, R., de Rooij, P. G. H., Kakembo, D., Bakker, P. A. H. M., Pieterse, C. M. J., Budko, N. V., & Berendsen, R. L. (2025). Seed tuber microbiome can predict growth potential of potato varieties. Nature Microbiology, 10(1), 28–40.