Comparing regression-based approaches for identifying microbial functional groups

Microbial communities are composed of functionally integrated taxa, and identifying which taxa contribute to a given ecosystem function is essential for predicting community behaviors. This study compares the effectiveness of a previously proposed method for identifying 'functional taxa,' ensemble quotient optimization (EQO), to a potentially simpler approach based on the least absolute shrinkage and selection operator (LASSO). In contrast to LASSO, EQO uses a binary prior on coefficients, assuming uniform contribution strength across taxa. Using synthetic datasets with increasingly realistic structure, we demonstrate that EQO's strong prior enables it to perform better in low-data regime. However, LASSO's flexibility and efficiency can make it preferable as data complexity increases. Our results detail the favorable conditions for EQO and emphasize LASSO as a viable alternative.