WAVDeSc: Wavelet Denoising for single-cell RNA-seq Data

Table of Contents

• Background
• Repository Structure
• Workflow
• Datasets
• Denoising Methods
• Usage
• Citation
• License
• Reference

Background

WAVDeSc leverages biorthogonal wavelet transforms to decompose scRNA-seq data into different frequency components and applies Bayesian thresholding to remove noise. The pipeline comprises three main phases: Signal Decomposition, Thresholding, and Signal Reconstruction aimed at producing a denoised scRNA-seq output. The approach enables the recovery of technical zeros and enhances quality while preserving important biological signals and improving downstream analyses.

---

Repository Structure

This repository contains the datasets and scripts used to generate the figures in our analysis of scRNA-seq denoising using (DCA, ENHANCE, MAGIC, SAVER) and WAVDESC the tool we propose.

├── Datasets
├── Denoising Methods
├── Evaluation Methods
├── WAVDeSc Module
├── Results
├── Scripts
└── READ ME

---

Workflow

The diagram below outlines the end-to-end workflow of the WAVDesC pipeline, highlighting its main processing steps and data flow from initial preprocessing to final output.

---

Datasets

All simulated and real datasets used and the denoised dataseted generated by the WAVDesC pipeline and the other denoising pipelines are stored in the datasets/ directory.*

• Real data
• Denoised data
• Simulated data

Slightly more descriptive:

• All input datasets and intermediate results generated during the WAVDesC workflow are organized in the data/ directory for easy access.

Do you want me to make it sound more formal (like a methods section) or practical (like a usage guide)?

---

Figures, Description, Corresponding Scripts and Data used

Figure	Description	Script	Data
Fig 3	Heatmaps	scripts/01_heatmap_script.R	Data
Fig 4	Coefficient of Variation Profiles	scripts/02_cv_profiles.ipynb	Data
Fig 5	Cell Clusters Plot	scripts/03_cell_clustering.R	Data
Fig 5	ARI_NMI Plot	scripts/03_ari_nmi.ipynb	Data
Fig 6	Performance Metrics Line Graph	scripts/04_DEGs_lingraph.ipynb	Data
Fig 7	Computational Runtime	scripts/05_computational_runtime.ipynb	Data

---

Denoising Methods

This directory contains implementations of multiple single-cell RNA-seq denoising algorithms, along with corresponding execution scripts for running them.

Each method is represented by two files:

• *_algorithm.* → The main script implementing the denoising algorithm.
• *_execute.* → The execution script (or helper file) showing how to run the algorithm with example inputs.

---

Available Methods

Method	Algorithm File	Execution File
DCA	DCA_algorithm.py	DCA_execute.txt
ENHANCE	ENHANCE_algorithm.py	ENHANCE_execute.txt
MAGIC	MAGIC_algorithm.py	MAGIC_execute.py
SAVER	SAVER_algorithm.r	SAVER_execute.r
WAVDeSc	WAVDeSc_algorithm.m	WAVDeSc_execute.m

---

File Extensions

• .py → Python scripts
• .r → R scripts
• .m → MATLAB scripts
• .txt → Example command-line instructions

---

Usage

Getting Started

Ensure that you have MATLAB installed with:

• Wavelet Toolbox:
  
  licence('test', 'Wavelet_Toolbox')

WAVDeSc Function

Clone the repository

git clone https://github.com/imensah/WAVDeSc.git
cd WAVDeSc
cd module
cp module </path/to/working/you_directory>  # replace with actual path

Quick Start

Basic Usage

% Load and denoise scRNA-seq data with default parameters
denoised_data = WAVDeSc('path/to/data.csv');

Complete Workflow Example

% 1. Load your scRNA-seq data (TSV/CSV format)
raw_data = readtable('expression_data.csv', 'FileType', 'text');

% 2. Extract components
gene_names = raw_data{:,1};
cell_names = raw_data.Properties.VariableNames(2:end);
expression_matrix = table2array(raw_data(:,2:end));

% 3. Denoise with WAVDeSc
denoised = WAVDeSc(expression_matrix, ...
    'Wavelet', 'bior2.6', ...
    'DecompositionLevel', 3, ...
    'Verbose', true, ...
    'SaveOutput', true, ...
    'OutputPath', 'denoised_output.csv');

Input Format

WAVDeSc accepts three input formats:

1. File path (CSV/TSV):

   denoised = WAVDeSc('data.csv');

2. Numeric matrix (genes × cells or cells × genes):

   denoised = WAVDeSc(expression_matrix);

3. MATLAB table:

   denoised = WAVDeSc(data_table);

Common Use Cases

1. Basic Denoising with Visualization

[denoised, metrics, params] = WAVDeSc('data.csv', ...
    'Verbose', true, ...
    'PlotResults', true);

2. Custom Wavelet and Thresholding

denoised = WAVDeSc(expression_matrix, ...
    'Wavelet', 'db8', ...
    'ThresholdRule', 'Soft', ...
    'DenoisingMethod', 'UniversalThreshold');

3. Performance Evaluation with Ground Truth

% If you have ground truth data for benchmarking
[denoised, metrics] = WAVDeSc(noisy_data, ...
    'GroundTruth', true_data, ...
    'ComputeMetrics', true, ...
    'PlotResults', true);

% View metrics
fprintf('RMSE: %.4f\n', metrics.RMSE);
fprintf('Correlation: %.4f\n', metrics.Correlation);
fprintf('SNR Improvement: %.2f dB\n', metrics.SNR_improvement_dB);

4. Batch Processing Multiple Files

% Process multiple datasets
files = {'dataset1.csv', 'dataset2.csv', 'dataset3.csv'};

for i = 1:length(files)
    fprintf('Processing %s...\n', files{i});
    denoised = WAVDeSc(files{i}, ...
        'SaveOutput', true, ...
        'OutputPath', sprintf('denoised_%d.csv', i), ...
        'Verbose', false);
end

Configuration Parameters

Essential Parameters

Parameter	Default	Options	Description
Wavelet	'db6'	'db4', 'db6', 'db8', 'bior2.6'	Wavelet function for decomposition
DecompositionLevel	'auto'	'auto' or integer (1-8)	Depth of wavelet decomposition
DenoisingMethod	'Bayes'	'Bayes', 'UniversalThreshold', 'Minimax'	Thresholding method
ThresholdRule	'Hard'	'Hard', 'Soft'	Type of thresholding

Optional Parameters

Parameter	Default	Description
Orientation	'auto'	Data orientation: 'auto', 'genes_rows', or 'genes_cols'
NoiseEstimate	'LevelDependent'	Noise estimation: 'LevelDependent' or 'LevelIndependent'
SaveOutput	false	Save denoised data to file
OutputPath	'./WAVDeSc_output.csv'	Output file path
Verbose	true	Display progress messages
PlotResults	false	Generate visualization plots
ComputeMetrics	false	Compute performance metrics
GroundTruth	[]	Ground truth data for evaluation

Troubleshooting

Memory Issues

If you encounter "array exceeds maximum array size" errors:

% Solution 1: Reduce decomposition level
denoised = WAVDeSc(data, 'DecompositionLevel', 2);

% Solution 2: Use simpler wavelet
denoised = WAVDeSc(data, 'Wavelet', 'db4', 'DecompositionLevel', 3);

% Solution 3: Process subset first for testing
subset = data(1:2000, :);
denoised_subset = WAVDeSc(subset, 'DecompositionLevel', 3);

File Reading Issues

If your CSV/TSV file has mixed text and numeric data:

% Prepare data manually
raw_data = readtable('data.tsv', 'FileType', 'text');
gene_names = raw_data{:,1};
cell_names = raw_data.Properties.VariableNames(2:end);
expression_matrix = table2array(raw_data(:,2:end));

% Then denoise
denoised = WAVDeSc(expression_matrix);

% Save with labels
output_table = array2table(denoised, ...
    'RowNames', gene_names, ...
    'VariableNames', cell_names);
writetable(output_table, 'denoised.csv', 'WriteRowNames', true);

Output Files

If SaveOutput is enabled, WAVDeSc generates:

• Denoised expression matrix: CSV file with denoised values
• Preserves gene names (row names) and cell IDs (column names)
• Same dimensions as input data

Tips for Best Results

1. Start with default parameters for initial testing
2. Use lower decomposition levels (2-4) for large datasets
3. Enable visualization (PlotResults) to inspect denoising quality
4. Test on a subset before processing entire dataset
5. Compare different wavelets (db4, db6, bior2.6) for your specific data
6. Use hard thresholding for sparse scRNA-seq data (default)
7. Save intermediate results for reproducibility

Citation

If you use WAVDeSc in your research, please cite:

@article{wavdesc2025,
  title={WAVDeSc: Wavelet-Based Denoising for Single-Cell RNA Sequencing Data},
  author={Mensah, Isabel and Appati, Justice Kwame and Salifu, Samson Pandam and Amoako-Yirenkyi, Peter},
  journal={In review - Array},
  year={xxxx}
}

License

• This project is open-source.

Reference

1. Poggi, J. M. (1996). Wavelet Toolbox: For Use with MATLAB. The MathWorks.