SIMBSIG: similarity search and clustering for biobank-scale data

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Modern bioinformatics applications often involve datasets too large for standard in-memory computation.
• Challenges arise in statistical genetics and single-cell analysis due to massive data sizes.

Purpose of the Study:

• Introduce SIMBSIG, a scalable Python package for out-of-core, GPU-enabled analysis.
• Provide a scikit-learn-like interface for similarity searches, PCA, and clustering.
• Facilitate efficient analysis of large biobank datasets.

Main Methods:

• Developed SIMBSIG with a PyTorch backend for modularity and GPU acceleration.
• Implemented out-of-core processing to handle datasets larger than available RAM.
• Designed for similarity searches, principal component analysis (PCA), and clustering.

Main Results:

• SIMBSIG offers a scalable solution for large-scale bioinformatics tasks.
• The package supports various data types, with a focus on biobank data.
• Achieves efficient computation through GPU integration and out-of-core capabilities.

Conclusions:

• SIMBSIG addresses the challenge of large-scale data analysis in bioinformatics.
• Its flexible design and GPU acceleration make it suitable for demanding applications.
• Enables advanced analyses like similarity search, PCA, and clustering on massive datasets.