ELAplus: Fast and accurate analysis platform for energy landscape analysis facilitated by fine-tuning optimization algorithm

Large-scale community-composition datasets, especially from microbiome studies, increasingly provide opportunities to identify major community compositional types (e.g., enterotypes in human microbiomes) and potential transitions depending on environmental factors. Energy landscape analysis based on maximum entropy models has emerged as a promising framework for characterizing such multi-stability in ecological communities. However, its application to diverse, high-dimensional compositional datasets remains limited by computational inefficiency, insufficient evaluation of predictability, and lack of systematic assessment of uncertainty.

Here, we present a computationally tractable inference framework for energy landscape analysis of multispecies communities, implemented in the R package ELAplus. We introduce a framework combining cross-validation-based selection of optimization settings, enabling accurate and computationally efficient model fitting across a wide range of simulated community datasets. In addition, we incorporate a bootstrap-based approach to quantify the reliability of inferred stable states, providing a systematic measure of uncertainty in landscape structures.

Simulation analyses demonstrate improved predictive performance and robustness compared to existing implementations. Applications to simulated occurrence datasets further illustrate how the framework can reveal stable states, basins of attraction, and potential tipping points under varying environmental conditions. The package also provides visualization tools, including disconnectivity graphs and energy surface plots, to facilitate intuitive interpretation of complex ecological landscapes.

Our framework enables robust and computationally efficient inference of ecological stability from compositional and environmental data, expanding the applicability of energy landscape approaches in diverse natural communities.