PIKE-R2P: Protein–protein interaction network-based knowledge embedding with graph neural network for single-cell RNA to protein prediction

Table 3 Impact of prior knowledge embedding on model performance of PIKE-R2P

	CBMC		PBMC
	PCC	MSE	PCC	MSE
No prior knowledge	0.8452 ± 0.0020	0.1960 ± 0.0022	0.8119 ± 0.0049	0.4432 ± 0.0043
Empirically determined interaction	0.8464 ± 0.0011	0.1958 ± 0.0018	0.8159 ± 0.0038	0.4306 ± 0.0073
Automated text mining	0.8456 ± 0.0011	0.1953 ± 0.0014	0.8165 ± 0.0012	0.4337 ± 0.0055
Database annotated	0.8460 ± 0.0031	0.1957 ± 0.0018	0.8163 ± 0.0030	0.4320 ± 0.0068
Combined score	0.8459 ± 0.0029	0.1952 ± 0.0060	0.8162 ± 0.0020	0.4333 ± 0.0072
Gene co-occurrence	0.8442 ± 0.0012	0.1944 ± 0.0027	0.8165 ± 0.0019	0.4329 ± 0.0039
Merge 5 features	0.8462 ± 0.0037	0.1944 ± 0.0035	0.8181 ± 0.0013	0.4303 ± 0.0083

The bold numbers represent the best performance. Note that on the CBMC dataset, for either PCC or MSE, the best and the second best scores are very close to each other, so both results are in bold

ISSN: 1471-2105