GoKawiil(a) The workflow of multi-voltage sweeping and machine learning. In brief, each peptide was measured using a multi-voltage (+60 mV, +100 mV and +140 mV) protocol to acquire its tPAL events. For each voltage, 200 events were collected per peptide to construct the dataset. For each peptide, nine features (ΔI, SD, skew, kurt, t off , median, IQR, spectral centroid and spectral entropy) were extracted from the events at each of the three voltages and subsequently concatenated to form a 27-dimensional feature matrix. Machine learning was performed with the Classification Learner toolbox of MATLAB, and model performance was evaluated by tenfold cross-validation. With the multi-voltage sweeping, the highest validation accuracy reached 98.0% with the quadratic SVM model. (b) Representative tPAL events of 20 XTRSC peptide variants acquired at +100 mV. Here, X represents each of the 20 proteinogenic amino acids. tPAL events of 20 XTRSC peptide variants acquired at other voltages are demonstrated in Supplementary Figs. 18–23. Specifically, CTRSC, HTRSC and PTRSC produce two types of nanopore events. The type-1 events of HTRSC and PTRSC, characterized by long dwell times, are attributed to N-terminal coordination, while the type-2 events arise from side-chain coordination. For CTRSC, the side chain of N-terminal cysteine may also participate in coordination along with the N-terminal amino group, giving rise to the secondary type of event. (c) The confusion matrix of peptide classification generated by the quadratic SVM model using a dataset from multi-voltage sweeping. (d) t-SNE visualization of 20 XTRSC peptide variants (n = 4,200). For HTRSC and PTRSC, only type-1 events were collected for plotting, owing to the infrequent occurrence of type-2 events.
Source data