A hyperparameter benchmark of VAE-based methods for scRNA-seq

ABSTRACT

This paper presents a systematic benchmark of architectural hyperparameters for variational autoencoder (VAE) methods in single-cell RNA-seq batch integration. The study compares scVI, MrVI, and LDVAE across diverse datasets and feature regimes. The findings provide practical recommendations for optimizing model performance and ensuring robust biological signal capture in single-cell analysis.