Back

CoLa-VAE: Cell-Cell Communication-aware Variational Autoencoder with Dynamic Graph Laplacian Constraints

Chen, Y.; Qi, C.; Fang, H.; Luan, F.; Zhang, Z.; Arya, S.; Wei, Z.

2026-03-31 bioinformatics
10.64898/2026.03.28.715052 bioRxiv
Show abstract

Intercellular communication is a fundamental driver of cellular identity and tissue homeostasis, yet current single-cell representation learning frameworks predominantly model cell state as a function of intrinsic gene expression, neglecting the extrinsic signaling context. Conversely, dedicated communication inference tools are often constrained by the sparsity and noise of raw transcriptomic data. Here, we present CoLa-VAE, a deep generative framework that explicitly integrates cell-cell communication (CCC) constraints into latent variable learning. By employing a dynamic graph Laplacian regularization derived from pairwise ligand-receptor interactions, CoLa-VAE disentangles communication-driven topology from intrinsic transcriptional heterogeneity within a variational autoencoder architecture. We demonstrate that CoLa-VAE serves as a robust, method-agnostic framework compatible with diverse signaling definitions, consistently outperforming state-of-the-art baselines in structural clustering metrics and denoising fidelity across heterogeneous sequencing platforms.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.