shiba2corr

Quantify the reproducibility of differentially spliced events between two Shiba experiments via a weighted Pearson correlation of dPSI values, computed on the union of differentially spliced events (DSEs) and weighted by the reliability of each PSI estimate.

Overview

Two Shiba runs (a Target condition and a Reference condition) produce per-event PSI tables. A naive Pearson correlation of their dPSI values overweights noisy events with low junction coverage. shiba2corr instead computes a weighted Pearson correlation on the union of DSEs from both runs, where the weight of each event reflects how precisely its PSI was estimated in each condition.

Weighted correlation:

$$r_w = \frac{\mathrm{Cov}_w(x, y)}{\sqrt{\mathrm{Var}_w(x)\ \mathrm{Var}_w(y)}}$$

with unbiased weighted estimators

$$\bar{x}_w = \frac{\sum w_i x_i}{\sum w_i},\quad \mathrm{Cov}_w(x,y) = \frac{\sum w_i (x_i - \bar{x}_w)(y_i - \bar{y}_w)}{\sum w_i - \frac{\sum w_i^2}{\sum w_i}}$$

and significance assessed via a t-test on the Kish effective sample size

$$n_{\text{eff}} = \frac{(\sum w_i)^2}{\sum w_i^2}.$$

In addition to the correlation table, shiba2corr produces:

• per-event-type KDE-coloured scatter plots of Target vs Reference dPSI;
• DSE overlap Venn diagrams (Target vs Reference, split by dPSI direction) for each event type and a combined grid.

Installation

pip install shiba2corr

For development:

git clone https://github.com/Sika-Zheng-Lab/shiba2corr.git
cd shiba2corr
pip install -e ".[dev]"

Usage

Quick start with example data

A small synthetic Shiba result pair is included in example/:

shiba2corr \
    -t example/target \
    -r example/reference \
    -o example/output

Basic usage

shiba2corr -t target_shiba_dir -r reference_shiba_dir -o output/

Filter DSEs by t-test P-value

If Shiba was run with t-tests enabled (column p_ttest present in the PSI files), restrict the analysis to events that also pass a P-value threshold:

shiba2corr -t target/ -r reference/ -o output/ --ttest 0.05

Choose a different weighting scheme

shiba2corr -t target/ -r reference/ -o output/ --weight-scheme geom_mean

Custom colors and font

shiba2corr -t target/ -r reference/ -o output/ \
    --target-color "#FF0000FF" \
    --reference-color "#00008BFF" \
    --font-family Arial

Skip Venn diagrams

shiba2corr -t target/ -r reference/ -o output/ --no-venn

Input File Format

-t / -r accept the path to a Shiba working directory. Within each directory, shiba2corr reads:

<shiba_dir>/results/splicing/PSI_SE.txt
<shiba_dir>/results/splicing/PSI_FIVE.txt
<shiba_dir>/results/splicing/PSI_THREE.txt
<shiba_dir>/results/splicing/PSI_MXE.txt
<shiba_dir>/results/splicing/PSI_RI.txt
<shiba_dir>/results/splicing/PSI_MSE.txt
<shiba_dir>/results/splicing/PSI_AFE.txt
<shiba_dir>/results/splicing/PSI_ALE.txt

Each PSI table is a tab-separated file with at least the following columns:

Missing event-type files are skipped with a warning.

-l/--event-list (optional)

A plain-text file with one pos_id per line. When provided, the weighted correlation is restricted to the intersection of the union DSE set and this list.

Output Files

Given -o output/:

TYPE is one of SE, FIVE, THREE, MXE, RI, MSE, AFE, ALE, or all (aggregated across event types). --no-venn skips all DSE overlap Venn output.

CLI Options

License

MIT License

Citation

If you use shiba2corr in your research, please cite this repository.

Contributing

Thank you for wanting to improve shiba2corr! If you have any bugs or questions, feel free to open an issue or pull request. See CONTRIBUTING.md for the development workflow and release process.

Authors

• Naoto Kubota (0000-0003-0612-2300)
• Sika Zheng (0000-0002-0573-4981)