Table 6.

(left) Accuracy of the semi-supervised block without the supervised loss. (right) Accuracy of the unsupervised block without the graph embeddings. Means and standard davations are computed over three different iterations

Dataset	w/o Predictor	w/o Graph Embeddings
BM-CITE	0.81 ± 0.002	0.48 ± 0.014
LUNG-CITE	0.88 ± 0.005	0.44 ± 0.008
PBMC-Multiome	0.94 ± 0.003	0.55 ± 0.071
PBMC-TEA	0.68 ± 0.007	0.69 ± 0.009
PBMC-DOGMA	0.87 ± 0.010	0.53 ± 0.041