Back to Models

SCALEX

✨ Generative Models

Online Single-Cell Data Integration

RNA

Scalable VAE for online integration of single-cell data by projecting into batch-invariant space

Publications

Online single-cell data integration through projecting heterogeneous datasets into a common cell-embedding space

Xiong et al.2022
Complexity
★★☆
moderate
Interpretability
★★★
high
Architecture
Batch-Aware VAE
Latent Dim
10
Used in LAIOR Framework

Online Batch Integration

SCALEX uses VAE to learn batch-invariant representations that enable continuous atlas expansion and online integration

Main Idea

Enable scalable data integration by learning disentangled batch and biological factors with reconstruction

Key Components

Batch-Invariant Encoder

Learns representations invariant to batch effects

Online Projection

Projects new data into existing latent space

Probabilistic Decoder

Reconstructs expression from latent codes

Mathematical Formulation

L = L_recon + KL(q(z|x)||p(z)) + L_batch

Loss Functions

ELBO
Reconstruction + KL divergence + Batch loss

Data Flow

Multi-batch Data → Encoder → Batch-Invariant Latent → Decoder → Integrated Expression

Architecture Details

Architecture Type

VAE with Batch Disentanglement

Input/Output Types

single-cell → reconstruction

Key Layers

BatchEncoderProjectionLayerDecoder

Frameworks

PyTorch

Tags

vaebatch-correctionintegrationrna