Fig. 10
From: A machine learning approach to predict metabolic pathway dynamics from time-series multiomics data
The ML approach can be used to produce biological insights. The left panel shows the final position in the proteomics phase space (similarly to the PCAP68 approach) for 50 strains generated by the ML algorithm by learning from the Michaelis–Menten kinetic model (Fig. 4) used as ground truth. Final limonene production is given by circle size and color. The PLS algorithm finds directions in the proteomics phase space that best align with increasing limonene production (component 1). Traveling in proteomics phase space along that direction (which involves overexpression of LS and underexpression of AtoB, PMD, and Idi, see Table S2) creates strains with higher limonene production. The ML approach not only produces biological insights to increase production but also predicts the expected concentration as a function of time for limonene and all other metabolites, generating hypotheses that can be experimentally tested (right panel)
