`MultiSampleAnalysis.filter_events`¶

Module: workflow.py

Flag noisy events and retain valid events. A hard blockage_lim filter is applied to the raw unitless blockade_ratio before the selected method.

Parameters¶

method (str): "blockade_gmm" by default; also supports "peak_detection", "isolation_forest", "lof", and "knn_background".
parameters (dict[str, Any] | None): Method-specific overrides.
blockade_gmm / peak_detection: blockade_col, dwell_col, rm_index, n_components=2, prior_mean=None, visualize=False.
isolation_forest / lof: contamination=0.05, feature_cols=["duration_s", "blockade_ratio", "segment_skew", "segment_kurt"].
knn_background: background_sample_ids, k=10, feature_cols, n_noise_match=1, background_ratio=1.0, metric="euclidean", random_state=42, and optional background_blockage_lim.
blockage_lim (tuple[float, float]): Hard blockade_ratio range, default (0.1, 1.0).

None. The method updates analysis.feature_df with is_noise and quality_tag, and stores valid rows in analysis.filtered_df.

Filtering is grouped by sample_id, falling back to trace_id.
knn_background requires background sample identifiers and compares target events with sampled background events in the chosen feature space.
Keep blockade_ratio on its raw ratio scale; do not pass a percentage such as (10, 80).