DBFE

DBFE extracts distribution-based features from genomic variant length distributions to enable machine learning analyses of variants derived from whole-genome sequencing data.

Key Features:

• Variant Length Binning: Categorizes genomic variants by length to produce size-range features from whole-genome sequencing variant calls.
• Clustering: Groups similar variants to derive distribution-based features reflecting variant pattern relationships.
• Density Estimation: Estimates variant distribution densities to quantify frequency patterns across the genome.
• Automation in Machine Learning Pipelines: Automates extraction of distribution-based features for integration into machine learning workflows.
• Validation on Cancer Whole-Genome Data: Validated on five real-world datasets comprising ovarian, lung, and breast cancer whole-genome genomes from 1,209 patients using four classification algorithms and a clustering approach.
• Unsupervised Learning Compatibility: Produces features that are applicable to unsupervised analyses of genomic samples without predefined labels.

Scientific Applications:

• Biomarker identification: Enables discovery of genomic fragments and variant distribution patterns associated with cancer.
• Prediction of treatment response: Facilitates machine learning models that associate variant distribution features with clinical responses to therapies.
• Cancer subtype classification: Supports classification of cancer subtypes from whole-genome variant-derived features.
• Exploratory genomic analysis: Supports unsupervised exploration of variant distributions across samples to reveal patterns without labels.

Methodology:

Computational feature extraction from whole-genome sequencing variant length distributions using variant length binning, clustering, and density estimation; automated integration into machine learning pipelines; validation on five ovarian, lung, and breast cancer whole-genome datasets totaling 1,209 patients using four classification algorithms and a clustering approach.