The rapid growth of ML in microbiome analysis presents challenges in interpretability and optimization. Our experience through the ML4Microbiome Cost Action emphasizes the importance of creating user-friendly tools with transparent standards to foster continued innovation. Contributions to open training material helped to utilize synergies between the COST action members and is creating long-lasting impact in disseminating methods that were evaluated and benchmarked as part of the ML4microbiome COST action. The DREAM challenge highlighted, for the first time, the relative performance of alternative modelling strategies in predicting future disease risk based on microbiome signatures. The independent benchmarking of the competing methods will support the development of novel prospective signatures not only for heart failure but potentially for other diseases. In addition, the analysis of the predictive microbiome features provides insights into the possible underlying mechanisms in microbiome-mediated disease associations. Data analysis poses various challenges, primarily stemming from the multitude of steps and methods involved, even for experts. We were able to produce an analyst-friendly summary of this process along with guidelines for it optimization, which we have published in several papers in a Special Issue in Frontiers Microbiology. In addition to having created a large and dynamic inter-disciplinary network of ML-microbiome scientists, we believe that these guidelines will contribute significantly to the establishment of microbiome analysis as a standard clinical practice to the benefit of human health.