Scyan

Scyan automates cell-type annotation in high-dimensional cytometry by using prior expert knowledge of the cytometry panel and a normalizing-flow deep generative model to map protein expression into a biologically meaningful latent space.

Key Features:

• Biological Knowledge Integration: Integrates prior expert knowledge of the cytometry panel to inform model-based annotation decisions.
• Deep Generative Model (Normalizing Flow): Employs a normalizing-flow deep generative model that maps protein expression into a biologically meaningful latent space for cell-type identification.
• Compatibility with Spectral Flow and Mass Cytometry: Operates on high-dimensional data produced by spectral flow and mass cytometers.
• Annotation without Manual Gating or Training Labels: Produces annotations using only prior panel knowledge, eliminating the need for manual gating or labelled training data.
• Batch-Effect Correction: Implements mechanisms to correct batch effects and improve reproducibility across experimental batches.
• Interpretability: Provides interpretable model outputs that enable inspection of annotation rationale.
• Complementary Tasks: Supports debarcoding to distinguish multiplexed samples and population discovery for identifying novel cell populations.
• Performance and Efficiency: Demonstrated faster and more accurate performance than prior models across multiple public datasets.

Scientific Applications:

• Single-cell phenotyping: Enables precise phenotyping of heterogeneous single-cell populations using high-dimensional cytometry data.
• Batch-aware comparative studies: Supports reproducible comparisons across experimental batches via batch-effect correction.
• Multiplexed sample analysis: Facilitates debarcoding and analysis of multiplexed cytometry datasets.
• Cell population discovery and quantification: Assists discovery, quantification, and characterization of known and novel cell populations.
• Domain-specific research: Applicable to studies in immunology, oncology, and developmental biology that rely on high-dimensional cytometric profiling.

Methodology:

Uses a normalizing-flow deep generative model to map protein expression into a biologically meaningful latent space, integrates prior expert knowledge of the cytometry panel, performs batch-effect correction, and supports debarcoding and population discovery without requiring manual gating or training labels.