Fig. 2: Optimized barcodes improve WarpDemuX performance.

a, b In silico barcode design. a Distinct wave-like signal patterns enhance the distinction between the individual target patterns. First, candidate barcodes are scored based on their similarity to a target pattern. Second, high-scoring candidates are selected, and an optimized set of barcodes is identified based on their mutual Dynamic Time Warping Distance (DTWD). b Identifying barcode sets using k-mer models. Identification of 12 barcodes with maximal pairwise DTWDs. c, d The use of optimized barcodes improves accuracy and yield. Values reported reflect the recall (confusion matrix, normalized over the actual class; true label), on both test sets combined, evaluated at a confidence cutoff of 0.5. c 4 optimized barcodes (WDX4) (d) 12 optimized barcodes (WDX12). e Relationship between accuracy and percentage of unclassified reads, modulated by the confidence cutoff. Values reflect the performance per model as the weighted average over their respective test sets. f Performance trends for accuracy and yield with the increasing number of barcodes, evaluated at a confidence cutoff of 0.5. Linear regression analysis of the 10-fold cross-validated performance (weighted averaged over test sets) suggests scalability for a larger set of optimized barcodes. Measured performance is shown by solid lines (mean ± std), and linear fits are represented by dashed lines, with annotated formulas. R² values are 0.75 (accuracy; blue) and 0.97 (yield; orange). Source data are provided as a Source Data file.