Flexynesis

Flexynesis integrates bulk multi-omics data using deep learning to facilitate precision oncology and clinical/pre-clinical data analysis.

Key Features:

• Unsupervised feature selection: Reduces dimensionality without requiring outcome labels.
• Multi-omics layer fusion options: Supports different strategies for combining heterogeneous molecular data types.
• Hyperparameter tuning: Optimizes model configuration automatically.
• Multiple neural network architectures: Includes fully connected networks, variational autoencoders, multi-triplet networks, graph neural networks, and cross-modality encoding networks.
• Standardized training and evaluation: Provides consistent model training, evaluation, and feature importance quantification methods.
• On-the-fly task determination: Automatically identifies whether the prediction task involves regression, classification, or survival modeling.
• Multi-task prediction: Predicts mixtures of numerical and categorical outcome variables with tolerance for missing labels.
• Structured data splitting: Enforces rigorous separation of data for model validation.

Scientific Applications:

• Precision oncology: Integrates genetic data, gene expression, protein abundance, and epigenetic measurements for clinical decision making.
• Clinical and pre-clinical data analysis: Enables deep learning-based analysis for researchers with or without deep learning development experience.
• Biomarker discovery: Supports identification of molecular markers through feature importance quantification and multi-omics integration.

Methodology:

Flexynesis processes bulk multi-omics data through deep learning architectures with structured data splitting, unsupervised feature selection, omics layer fusion, and hyperparameter tuning, followed by standardized training, evaluation, and feature importance quantification.