Fig. 5.

Evaluating 140 AI-generated cRBSs with desired TFB dynamic ranges and improving prediction accuracy. (A–E) Prediction accuracy analyses of the BAGAN-GP model for 28 synthetic glucarate biosensors in sub-libraries I–V, respectively. We suggested that the synthetic TFB was constructed by generated cRBSs in sub-library V, whose dynamic range was more extensive than trained ones (247-fold [7]). (F) Correlation analysis between desired and experimentally-observed dynamic ranges of TFBs controlled by BAGAN-GP-generated cRBSs in each sub-library. (G) The improved prediction accuracy of TFB dynamic ranges in sub-libraries I and II using the constructed filter model. DR: dynamic range. GS and CS are RBSs of the reporter sfGFP and regulator CdaR, respectively. The dotted box represents cRBSs whose desired and observed TFB dynamic ranges had a± 30% error margin. Data represent the mean and standard deviation of two replicates.