Research - DropDAE: Denosing Autoencoder with Contrastive Learning for Addressing Dropout Events in scRNA-seq Data
31 Jul 2025
Reading time ~1 minute
Project Overview
We developed a method DropDAE (Dropout Denoising Autoencoder), a novel deep learning framework that improves imputation for high-dimensional data like single-cell RNA sequencing (scRNA-seq). It uses:
- A denoising autoencoder structure to reconstruct corrupted input data.
- An optional triplet loss based on consensus clustering to improve separation between similar and dissimilar cells.
R package
We also developed a R package for DropDAE method. View R package
Contact
For more information about the paper or R package, please contact me.