SAUCIE

SAUCIE performs deep-learning-based analysis of single-cell data to correct batch effects, denoise measurements, produce low-dimensional visualizations, and cluster cells for biological signal extraction in large datasets such as mass cytometry.

Key Features:

• Deep Neural Network Architecture: SAUCIE employs a deep neural network with multiple hidden layers that enables scalable, parallelizable integration of complex single-cell data representations.
• Interpretable Feature Learning: The model incorporates regularization penalties to promote interpretability of features learned in hidden layers.
• Batch Correction and Denoising: The network performs batch correction and denoising across samples and experimental conditions while preserving biological signals.
• Low-Dimensional Visualization and Clustering: SAUCIE generates low-dimensional embeddings and performs unsupervised clustering to reveal cellular patterns and structures.

Scientific Applications:

• Large-scale immunophenotyping (mass cytometry): Applied to 11 million T cells from dengue patients in India measured by mass cytometry, SAUCIE batch corrected the data and identified cluster-based signatures indicative of acute dengue infection.
• Patient-level immune stratification: The method facilitated creation of a patient manifold that stratified immune responses to dengue.

Methodology:

Input data are processed through multiple hidden layers of a deep neural network that perform batch correction, denoising, and feature extraction simultaneously, producing corrected data, low-dimensional embeddings, and clusters.