Fig. 4: Developing an optimized nanopore barcode set.

a Confusion matrix showing results of ResNet-18 CNN trained on all 36 characterized barcodes. CNN achieved an accuracy of ~67% on the test set. b Density plot showing the distribution of mean fractional current for each barcode selected for the highly separable set. Each distribution is composed of ~13000 data points. c Confusion matrix showing results of ResNet-18 CNN trained only on the selected highly separable barcodes, which achieved an accuracy of 93% on the test set. d Bar plots comparing normalized capture frequencies of each circuit output in two multiplexed samples. Circuits 1, 5, 7, 8, and 9 (with Barcodes B10, B7, C13, C8, and C12, respectively, see Supplementary Tables 1 and 2) were present in each sample. In the left plot, inputs for Circuits 0 and 5 were added. In the right plot, inputs for Circuits 1, 7, and 8 were added. Each gate complex was present at 0.2 uM, input at 0.2 uM, and streptavidin at 3.2 uM. Error bars represent ± standard deviation of three biological replicates.