Skip to contents

Detect outliers (peaks and valleys) in ONT signal using z-scores

Source: R/ninetails_core_functions.R
filter_signal_by_threshold.Rd

Detects areas in the poly(A) tail signal where values significantly deviate from those typical of an adenosine homopolymer. The function produces a vector of pseudomoves with values in the range {-1, 0, 1}, where -1 corresponds to signals significantly below the local mean (potential C/U), 1 corresponds to signals significantly above the local mean (potential G), and 0 corresponds to typical adenosine homopolymer values.

Usage

filter_signal_by_threshold(signal)

Arguments

signal

Numeric vector. An ONT read fragment corresponding to the poly(A) tail region of the read of interest, as delimited by nanopolish polya function. Fragments are stored in tail_feature_list[[1]][[readname]][[2]] produced by create_tail_feature_list.

Value

Numeric vector of pseudomoves corresponding to the analyzed tail region, with the same length as the input signal. Values are integers in the range {-1, 0, 1}:

-1

Signal valleys (potential C/U nucleotides)

0

Normal signal (likely A nucleotides)

1

Signal peaks (potential G nucleotides)

Details

The pseudomoves vector allows more accurate calibration of nucleotide positions of potential non-adenosine residues than the moves produced by the Guppy basecaller.

The algorithm implements a sliding-window approach based on adaptive z-scores:

  1. A calibration phase (100 synthetic data points drawn from the most frequent signal values) establishes baseline signal characteristics

  2. Rolling mean and standard deviation are computed over a 100-point adaptive window

  3. Outliers are detected when signal deviates > 3.5 standard deviations from the local mean

  4. Direction of deviation determines the pseudomove value

Terminal positions (first 5) are masked to prevent false positives at the adapter-tail boundary. This is a temporary safeguard until the resegmentation model is refined.

References

Based on: Brakel, J.P.G. van (2014). "Robust peak detection algorithm using z-scores". Stack Overflow. Available at: https://stackoverflow.com/questions/22583391/peak-signal-detection-in-realtime-timeseries-data/22640362#22640362 (version: 2020-11-08).

See also

substitute_gaps for gap handling, create_tail_feature_list for the complete feature extraction pipeline

Examples

if (FALSE) { # \dontrun{

ninetails::filter_signal_by_threshold(
  signal = tail_feature_list[[1]][["readname"]][[4]]
)

} # }