ACTINN

ACTINN applies neural network classification to automate cell type identification in single-cell RNA sequencing (scRNA-seq) data using a model with three hidden layers to learn gene expression patterns for cross-dataset prediction.

Key Features:

• Neural Network Architecture: A neural network with three hidden layers is trained on datasets with predefined cell type annotations to learn complex patterns in gene expression data.
• Training Datasets: The model was trained on the Tabula Muris Atlas (mouse) and a human immune cell dataset.
• Application Scope: ACTINN has been applied to predict cell types in mouse leukocytes, human peripheral blood mononuclear cells (PBMCs), and human T cell subtypes.
• Speed and Accuracy: The model is designed for rapid processing and to mitigate unwanted variation and the absence of canonical markers that can affect clustering-based methods.
• Complementary Tool: ACTINN functions alongside existing scRNA-seq methodologies to reduce manual annotation and biases from literature-based lookup.
• Software Implementation: The codebase is implemented in Python.

Scientific Applications:

• Cell Type Identification: Automated classification of cell types in scRNA-seq datasets for large-scale studies.
• Cross-Species Analysis: Application to both mouse and human datasets supports comparative and translational research.
• Research Efficiency: Reduction of manual annotation and literature lookup accelerates annotation of large or complex datasets.

Methodology:

Train a neural network with three hidden layers on datasets with known cell type annotations (e.g., Tabula Muris Atlas and a human immune cell dataset), then apply the trained model to predict cell types for new scRNA-seq datasets based on learned parameters to reduce noise and variability effects.