Analysis workflow

The API and local UI use the same event-centered workflow. The detector-tuning step is a preview on a selected local interval; full event detection runs separately over the loaded traces.

Stage	Purpose	Main outputs
Import	Read ABF or CSV current traces and assign sample groups	Traces and sample metadata
Preprocess	Denoise signals and inspect raw or processed traces	Processed traces
Pre-Events	Tune event-detection settings on a short time window	Preview events and detector settings
Events	Detect events across full traces	Events and baseline information
Features & Filter	Calculate event features and remove low-quality or outlier events	Feature table and filtered events
Reduction	Explore numeric event features with PCA, t-SNE, or UMAP	Coordinates appended to the event table
Train Model	Fit classical ML or waveform DL models	Model package and evaluation results
Predict	Classify events in new samples with a trained model	Per-event labels and scores
Export	Save selected tables, parameters, and analysis outputs	Files in the chosen export directory

Event data and labels

Each trace is associated with a sample ID. sample_to_group maps sample IDs to experimental groups; feature extraction uses these group values as event labels. Without that mapping, add a label column before supervised training.

The default classical-model features are duration_s, blockade_ratio, segment_std, segment_skew, and segment_kurt. blockade_ratio is a unitless ratio. The common display and filtering range is 0–1, but out-of-range values should be checked rather than silently rescaled.

Detection and filtering

The Python API supports threshold, zscore_threshold, cusum, pelt, and hmm event detection, along with rolling or global baseline options. Set detector parameters explicitly when reproducing a run; UI and API defaults can differ. See event detection reference and preview detection reference.

Filtering first applies blockage_lim, then the selected quality or anomaly method, such as blockade_gmm, peak_detection, isolation_forest, lof, or knn_background.