CAMLU

CAMLU employs an autoencoder-based supervised learning framework to annotate cell types in single-cell RNA sequencing (scRNA-seq) datasets and to detect novel cell types via reconstruction errors.

Key Features:

• Autoencoder Integration: Employs an autoencoder trained on labeled data to reconstruct input sequences and compute reconstruction errors on testing datasets.
• Iterative Feature Selection: Implements iterative feature selection that identifies features exhibiting a bi-modal pattern in their distribution to refine analyses.
• Novel Cell Type Identification: Uses autoencoder reconstruction errors to pinpoint cells absent from the training set, reducing the number of unlabeled cells.
• Support Vector Machine (SVM) Integration: Integrates a support vector machine for classification and annotation across detected cell types.
• Validation on Real scRNA-seq Data: Demonstrated performance through numerical experiments on five real scRNA-seq datasets showing superior results compared to existing methods.

Scientific Applications:

• Developmental Biology: Enables precise annotation of cell types to study developmental trajectories and tissue differentiation.
• Immunology: Facilitates identification of known and novel immune cell subtypes in heterogeneous immune populations.
• Oncology: Supports characterization of tumor heterogeneity by detecting distinct and novel tumor-associated cell types.

Methodology:

Trains an autoencoder on labeled scRNA-seq training data, reconstructs inputs and computes reconstruction errors on testing data, performs iterative feature selection by identifying bi-modal features, uses reconstruction errors to flag novel cells and reduce unlabeled cells, and integrates a support vector machine for classification.