Denoising autoencoder-based soft K-medoid clustering for single-cell RNA sequence data

Single-cell RNA sequencing enables comprehensive exploration of cellular diversity, yet clustering this data poses significant computational challenges due to its scale.Existing deep learning clustering algorithms for single cells fail to learn cluster-friendly lowdimensional space due to ignoring pairwise distance constraints or making additional latent space assumptions like mixture Gaussian distribution.Therefore, in this work to evaluate scRNA-seq data, a "deep learning" method is merged with a denoising autoencoder.A soft K-medoid approach is also presented to group the cell population according to the learned latent space.It is termed as "scCskm", which stands for single cell clustering with soft K-medoids method.This approach had an NMI and ARI score of 0.77 and 0.75 respectively, outperforming most of the other on-market clustering algorithms.Also, scCskm is comaparatively fast, significantly reducing processing load during preprocessing and clustering phase compared to most of the other methods.