splicespectrax 2025.0.3

Splicing-based Pattern Extraction and Clustering using TRAnscriptomics

SPECTRA: Unsupervised Analysis of Alternative Splicing

✂️ About

SPECTRA (Splicing-based Pattern Extraction and Clustering using TRAnscriptomics) is an end-to-end pipeline for discovering patient subtypes based on alternative splicing. It is a modernized and optimized version of the original OncoSplice algorithm.

SPECTRA leverages an iterative clustering strategy to identify stable and dominant splicing patterns across patient samples. The new implementation enhances the speed, accuracy, and modularity, allowing seamless integration into both command-line workflows and interactive Python environments.

📌 Installation

SPECTRA can installed as a Python package via pip. We recommend using conda environment with python version set to 3.12.

pip install -i https://test.pypi.org/simple/ spectra-kairaveet

📚 Documentation

Detailed documentation on how to perform SPECTRA analysis is provided on https://spectra-kairaveethakkar.readthedocs.io/en/latest/ now.

👩‍🏫 Tutorial

Example datasets: PSI files for each TCGA cancer can be downloaded from https://www.synapse.org/Synapse:syn64934289.

SPECTRA can be used in two ways:

• As a command-line tool for end-to-end execution
• As a modular workflow, where individual functions are called step-by-step

See the tutorials and example scripts for each approach:

Command-Line Interface (CLI)

Run the entire pipeline with a single command using main.py. This is ideal for multiple dataset processing and automated workflows.

Modular Usage

Import and run individual components such as preprocessing, clustering, or visualization in a custom step-by-step analysis.

📝 Overview of Modules

Module	Description
main.py	Entry point for running the complete SPECTRA pipeline. Handles argument parsing and execution flow.
round_wrapper.py	Wraps a single iteration of clustering (SPECTRA performs 3 iterations by default).
preprocess.py	Performs variance-based and intercorrelation-based filtering of splicing events prior to clustering.
remove_redundancy.py	Removes redundant splicing events based on intra-gene correlation.
feature_selection.py	Implements PCA-based feature selection, similar to the splice-ICGS method in the original OncoSplice.
median_impute.py	Imputes missing values in the splicing matrix using the median for each event.
visualizations.py	Generates visual summaries, including splicing event annotation bar plots and cluster heatmaps.
determine_rank.py	Automatically determines the optimal NMF rank (if not user-specified).
run_nmf.py	Performs NMF clustering and assigns multi-label cluster memberships.
metadata_analysis.py	Analyzes and annotates differential splicing events across clusters.
linear_svm.py	Applies linear SVM for final cluster assignment.
correlation_depletion.py	Identifies and depletes splicing events associated with a clustering round.
correlation_depletion_vectorized.py	A faster version of correlationDepletion.py using imputed values and optimized calculations.

📖 Citation

Coming soon — citation information for referencing SPECTRA in publications.