Ponente
Descripción
Microbial communities exhibit complex eco-evolutionary dynamics shaped by microbe–microbe and microbe–environment interactions. However, reconstructing and interpreting microbial interaction networks remains a major challenge due to the limitations of correlation-based approaches and the high dimensionality of microbiome data. Here, we combine synthetic community experiments, metagenomic data analysis, Bayesian inference, and graph learning to investigate microbial interactions across multiple ecosystems.
To improve the interpretation of ecological networks, we implemented and trained a Graph Neural Network to identify structurally important species from pairwise interaction networks. Across simulated datasets, network topology consistently outperformed interaction strength and standard graph statistics as a predictor of species importance.
We then applied complementary experimental and computational approaches across distinct microbial ecosystems. In rhizosphere synthetic communities, we systematically varied bacterial composition and temperature to directly estimate bacteria–bacteria interactions measured as net fitness effects. We found that community composition had a stronger overall effect than temperature, but that temperature sensitivity was strongly associated with network connectivity.
In ocean microbial communities, we reconstructed interaction networks from global metagenomic datasets using a molecularly informed framework linking metagenome-assembled genomes to biosynthetic gene clusters. This approach reduced spurious associations and revealed that secondary metabolites associated with cooperative “common good” functions are strongly enriched as network hubs in the photic ocean.
Finally, to quantify microbial responses to climate change, we evaluated synthetic communities derived from the Cuatro Ciénegas oasis, in northern Mexico, under multiple warming scenarios. Preliminary results show that compositionally similar communities can exhibit dramatically different responses to temperature, suggesting that higher-order interactions drive context-dependent ecosystem responses to warming.
Together, these results demonstrate how integrating experimental microbiology with quantitative network inference and machine learning can improve our understanding of microbial community organization and resilience under environmental change.