Fig. 4: Second-round optimization of the E. coli extract-based CFE system.
From: AI-driven high-throughput droplet screening of cell-free gene expression
a, b In-droplet screening to optimize the concentrations of the selected components. a The fluorescence coding strategy. b A 3D scatter plot showing the fluorescence intensity profile of the droplets. The color of the dots indicates the sfGFP intensity. c–e In silico optimization. c An extreme gradient boosting model used in the optimization. d Concentration ranges scanned in the model prediction. e The estimation matrix sorted by yield level. f–i In vitro verification. f Fold change in the expression yield of 12 proteins with the simplified and optimized CFE formulation compared to the original recipe. The two sfGFP columns depict the same data because sfGFP was considered fully dissolvable (see “Methods” for fluorescence-based quantification of sfGFP), and the data are shown as mean ± s.d. of three independent experiments. Western blot results of (g) Sfp (27 kDa) and (h) CAR (128 kDa). i SDS-PAGE analysis of Vlm1 (370 kDa). M, T, and S are marker, total fraction, and soluble fraction, respectively. The gel images are representative examples from three independent experiments, all yielding consistent results. The background shadings highlight a representative round of DropAI optimization: in-droplet screening (clear), in silico optimization (grey), and in vitro verification (green).
