scMAE: a masked autoencoder for single-cell RNA-seq clustering

scMAE: a masked autoencoder for single-cell RNA-seq clustering

16 January 2024 | Zhaoyu Fang, Ruiqing Zheng, Min Li
The paper introduces scMAE, a novel masked autoencoder-based method for single-cell RNA sequencing (scRNA-seq) data clustering. scMAE addresses the challenge of effectively capturing gene correlations by perturbing gene expression data and using a masked autoencoder to reconstruct the original data. The masked autoencoder includes a masking predictor that predicts whether gene expression values are masked, enhancing the model's ability to capture latent structures and dependencies in the data. Extensive experiments on 15 scRNA-seq datasets from various platforms show that scMAE outperforms state-of-the-art methods in terms of clustering performance, particularly in identifying rare cell types. Biological analyses confirm the significance of the identified cell subpopulations. The source code for scMAE is available at <https://zenodo.org/records/10465991>.The paper introduces scMAE, a novel masked autoencoder-based method for single-cell RNA sequencing (scRNA-seq) data clustering. scMAE addresses the challenge of effectively capturing gene correlations by perturbing gene expression data and using a masked autoencoder to reconstruct the original data. The masked autoencoder includes a masking predictor that predicts whether gene expression values are masked, enhancing the model's ability to capture latent structures and dependencies in the data. Extensive experiments on 15 scRNA-seq datasets from various platforms show that scMAE outperforms state-of-the-art methods in terms of clustering performance, particularly in identifying rare cell types. Biological analyses confirm the significance of the identified cell subpopulations. The source code for scMAE is available at <https://zenodo.org/records/10465991>.
Reach us at info@study.space