Figure 1.

Schematic illustration on the discover–model–learn–advance (DMLA) cycle for microbiome technology development. In this study, we discovered that light wavelengths can be utilized to bidirectionally regulate bio‐denitrification to nitrogen gas or nitrite for different nitrate removal strategies. After that, we conducted metatranscriptomic sequencing and data preprocessing to obtain graph‐structured data sets for modeling. Graph neural networks, a representative geometric deep‐learning approach, were utilized to unsupervisedly learn the gene panels. On the basis of the critical gene panels, we learned the coexpressed pathways and genes through the model toolkits we proposed and validated the knowledge we learned in the wet lab, which drove the biotechnology advancement, including new applications and mechanism discovery. The new mechanism deepened our recognitions on optogenetics in microbiomes that wavelength‐dependent secretion systems played a pivotal role in the collective behavior of microbiota in response to light wavelength. The secreted active substances and proteins mediated the cross‐cellular interactions.