Fig. 5: Demonstration level prediction and tuning of P. somniferum NMCH and CPR for improved reticuline production from norcoclaurine.
From: Machine learning discovery of missing links that mediate alternative branches to plant alkaloids
a Comparison of active site configurations of positive training example EcNMCH (blue) with that of candidate sequence PsNMCH Isoform 1 (PsNMCH-I1, green). b For visual representation, two-dimensional SVM-based prediction of NMCH sequences is shown, with positive and negative prediction spaces colored green and white, respectively. Selected sequences are shown in bold. With exception to EcNMCH and EcNMCH-Y202H (brown), all points represent P. somniferum sequences. PCA is used to compress multi-dimensional data into two dimensions (PC1 and PC2) for the visual representation. Corresponding high-dimensional SVM results are detailed in Supplementary Table 4, and high-dimensional SVM decision scores are listed. c Two-dimensional SVM-based prediction of CPR sequences, with positive and negative prediction spaces colored green and white, respectively. With exception to positive training sequence AtATR2 (blue), all points represent tested P. somniferum sequences. PCA is used to compress multi-dimensional data into two dimensions (PC1 and PC2) for the visual representation. Corresponding high-dimensional SVM results are detailed in Supplementary Table 5, and high-dimensional SVM decision scores are listed. d Conversion of 1.2 mM norcoclaurine to reticuline, mediated by various combinations of NMCH and CPR, together with Ps6OMT, PsCNMT and Ps4OMT, in strains N1-01-DE3, N1-02-DE3, N1-03-DE3, N1-04-DE3, N2-01-DE3, N2-02-DE3, N2-03-DE3, N2-04-DE3 (Supplementary Table 2). Here, individual samples were analyzed 3 times (n = 3) to generate bar graphs in Prism 7 version 7.0d with error bars representing mean values +/− standard deviations. Source data are provided as a Source Data file.
