Table 2 Table showing the accuracy of K-NN classifiers trained on the latent representation of different methods to predict cell types in the five datasets. The KNN classifier is trained on 50% of the cells and tested on the rest with k=10. The standard deviation is computed by training 5 KNN-classifiers on randomly chosen cells for each experiment
From: SHARE-Topic: Bayesian interpretable modeling of single-cell multi-omic data
| Â | Mouse brain | Mouse skin | B-lymphoma | Pbmc10k | Mouse cortex |
|---|---|---|---|---|---|
MOFA+ | 0.479 \(\pm 10^{-2}\) | 0.379 \(\pm 2\times 10^{-2}\) | 0.813 \(\pm 3\times 10^{-3}\) | 0.802 \(\pm 5\times 10^{-3}\) | 0.66 \(\pm 5\times 10^{-3}\) |
scGlue | 0.803 \(\pm 10^{-2}\) | 0.834 \(\pm 2\times 10^{-3}\) | 0.894 \(\pm 2\times 10^{-3}\) | 0.895 \(\pm 3\times 10^{-3}\) | 0.853 \(\pm 4\times 10^{-3}\) |
Seurat (PCA, LSI) | 0.854 \(\pm 8\times 10^{-3}\) | 0.887 \(\pm 2\times 10^{-3}\) | 0.904\(\pm 2\times 10^{-3}\) | 0.896 \(\pm 5\times 10^{-3}\) | 0.863 \(\pm 4\times 10^{-3}\) |
SHARE-Topic | 0.830 \(\pm 10^{-2}\) | 0.754 \(\pm 3\times 10^{-3}\) | 0.871 \(\pm 10^{-3}\) | 0.880 \(\pm 2\times 10^{-3}\) | 0.756 \(\pm 7\times 10^{-3}\) |