predictCondition: Predict Condition

In Bishop-Laboratory/RLSeq: RLSeq: An analysis package for R-loop mapping data

Predict Condition

Description

Uses the results of analyzeRLFS to predict whether a sample is "POS" (robust R-loop mapping) or "NEG" (poor R-loop mapping). See details.

Usage

predictCondition(object, rlfsRes = NULL, ...)

Arguments

object	An RLRanges object with analyzeRLFS already run. Ignored if rlfsRes is provided.
rlfsRes	If object not supplied, provide the rlfsRes list which is obtained from rlresult(object, "rlfsRes").
...	Internal use only.

Details

Following R-loop forming sequences (RLFS) analysis, the quality model (see RLHub::models) is implemented for predicting the sample condition in coordination with other results from analyzeRLFS. A prediction of “POS” indicates robust R-loop mapping, whereas “NEG” indicates poor R-loop mapping. The succeeding sections describe this process in greater detail.

Application of binary classification model

First, the binary classifier is applied, yielding a preliminary prediction of quality. This is accomplished via the following steps:

1. Calculate the Fourier transform of the Z-score distribution (see analyzeRLFS).

2. Reduce the dimensions to the engineered feature set (see table below).

3. Apply the preprocessing model (see RLHub::models) to normalize these features

4. Apply the classifier (see RLHub::models) to render a quality prediction.

Engineered feature set

Abbreviations: Z, Z-score distribution; ACF, autocorrelation function; FT, Fourier Transform.

feature	description
Z1	mean of Z
Z2	variance of Z
Zacf1	mean of Z ACF
Zacf2	variance of Z ACF
ReW1	mean of FT of Z (real part)
ReW2	variance of FT of Z (real part)
ImW1	mean of FT of Z (imaginary part)
ImW2	variance of FT of Z (imaginary part)
ReWacf1	mean of FT of Z ACF (real part)
ReWacf2	variance of FT of Z ACF (real part)
ImWacf1	mean of FT of Z ACF (imaginary part)
ImWacf2	variance of FT of Z ACF (imaginary part)

Final quality prediction

The results from the binary classifier are combined with other results from analyzeRLFS to yield a final prediction. To yield a prediction of “POS” all the following must be TRUE:

1. The RLFS Permutation test P value is significant (p < .05). Stored as PVal Significant in the results object.

2. The Z-score distribution at 0bp is > 0. Stored as ZApex > 0 in the results object.

3. The Z-score distribution at 0bp is > the start and the end. Sored as ZApex > ZEdges in the results object.

4. binary The classifier predicts a label of “POS”. Stored as Predicted 'POS' in the results object.

Value

An RLRanges object with predictions accessible via rlresult(object, "predictRes").

Structure

The results object is a named list of the structure:

• Features

  • A tbl with three columns that describe the engineered features used for prediction:

    • feature: the name of the feature (see details).

    • raw_value: The raw value of that feature in the supplied object.

    • processed_value: The normalized value of that feature after preprocessing (see details).

• Criteria

  • The four criteria which must all be TRUE to render a prediction of "POS" (see details).

• prediction

  • The final prediction. "POS" indicates robust R-loop mapping, "NEG" indicates poor R-loop mapping.

Examples

# Example data with analyzeRLFS already run
rlr <- readRDS(system.file("extdata", "rlrsmall.rds", package = "RLSeq"))

# predict condition
rlr <- predictCondition(rlr)

# With rlfsRes
predRes <- predictCondition(rlfsRes = rlresult(rlr, "rlfsRes"))

Bishop-Laboratory/RLSeq documentation built on Jan. 28, 2023, 11:38 p.m.