Scaden

Scaden infers cellular composition from bulk RNA-seq and microarray gene expression data using a deep neural network trained on single-cell RNA-seq to perform cell type deconvolution.

Key Features:

• Deep neural network architecture: Uses a deep-learning model to perform deconvolution from gene expression profiles.
• Training on single-cell RNA-seq: Models are trained using single-cell RNA sequencing datasets to learn cell-type-specific signals.
• Inference of cellular composition: Infers cell type proportions from bulk RNA-seq and microarray expression data.
• Discriminative feature learning: Generates discriminative features that improve robustness against bias and noise in expression data.
• Reduced preprocessing requirements: Mitigates the need for complex data preprocessing and manual feature selection typically used in traditional deconvolution methods.
• Multi-dataset integration: Leverages combined information from multiple datasets to enhance model performance.
• Cross-species applicability: Applicable to human and mouse tissue expression data.
• Performance characteristics: Demonstrates improved precision and robustness relative to existing cell deconvolution algorithms.

Scientific Applications:

• Cell type deconvolution: Determining cell type proportions from bulk RNA-seq and microarray datasets.
• Tissue composition analysis: Elucidating cellular compositions within tissue expression profiles across human and mouse samples.
• Integrative dataset analysis: Combining multiple datasets to improve accuracy of cellular composition estimates.
• Biological insight generation: Supporting investigations into developmental processes and disease mechanisms via inferred cellular composition.

Methodology:

A deep neural network is trained on single-cell RNA-seq datasets using gene expression data to learn discriminative features that enhance robustness to bias and noise, and models incorporate combined information from multiple datasets while reducing reliance on complex preprocessing and manual feature selection.