Speaker
Description
Combining RNAseq data and models in microbial ecology aims to reveal species interactions and improve health outcomes. However, data is often noisy, and inferred "interactions" are merely model-dependent correlations rather than direct biological mechanisms. This raises a crucial question: when inferring models from data, are complex models better, or is simplicity more effective?
We argue that minimal models generally outperform complex ones. Using information geometry and Bayesian inference, we demonstrate that simple models maximize reliable information extraction, making them information-theoretically optimal \cite{castro2025scarce}. Furthermore, many widely reported microbial macroecological patterns may simply result from data aggregation \cite{castro2026} or lack the robustness required to be genuine laws \cite{camacho2025microbial}.
Bibliography
@article{castro2025scarce,
title={Scarce data, noisy inferences and overfitting: the hidden flaws in ecological dynamics modelling},
author={Castro, Mario and Vida, Rafael and Galeano, Javier and Cuesta, Jose A},
journal={Journal of the Royal Society Interface},
volume={22},
number={231},
year={2025},
publisher={The Royal Society}
}
@article{castro2026,
title={In preparation},
author={Castro, Mario and Cuesta, Jose A},
}
@article{camacho2025microbial,
title={Microbial populations hardly ever grow logistically and never sublinearly},
author={Camacho-Mateu, Jos{\'e} and Lampo, Aniello and Castro, Mario and Cuesta, Jos{\'e} A},
journal={Physical Review E},
volume={111},
number={4},
pages={044404},
year={2025},
publisher={APS}
}
