Mathematical models of the colonic microbiota: an evaluation of accuracy using in vitro fecal fermentation data

Traditional approaches for studying diet-colonic microbiota interactions are time-consuming, resource-intensive, and often hindered by technical and ethical concerns. Metagenome-scale community metabolic models show promise as complementary tools to overcome these limitations. However, their experimental validation is challenging, and their accuracy in predicting colonic microbial function under realistic dietary conditions remains unclear. This study assessed the accuracy of the Microbial Community model (MICOM) in predicting major short-chain fatty acid (SCFA) production by the colonic microbiota of weaning infants, using fecal samples as a proxy. Model predictions were compared with experimental SCFA production using in vitro fecal fermentation data at the genus level. The model exhibited overall poor accuracy, with only a weak, significant correlation between measured and predicted acetate production (r = 0.17, p = 0.03). However, agreement between predicted and measured SCFA production improved for samples primarily composed of plant-based foods: acetate exhibited a moderate positive correlation (r = 0.31, p = 0.005), and butyrate a trend toward a weak positive correlation (r = 0.21, p = 0.06). These findings suggest that the model is better suited for predicting the influence of complex carbohydrates on the colonic microbiota than for other dietary compounds. Our study demonstrates that, given current limitations, modeling approaches for diet-colonic microbiota interactions should complement rather than replace traditional experimental methods. Further refinement of computational models for microbial communities is essential to advance research on dietary compound-colonic microbiota interactions in weaning infants.