Integrating pathway knowledge with deep neural networks to reduce the dimensionality in single-cell RNA-seq data

Table 4 Average retrieval performance across the different cell type

Architecture	Number of nodes (2nd layer)	HSC	4cell	ICM	Spleen	8cell	Neuron	Zygote	2cell	ESC	Mean
PCA 100 (with full gene space)	–	0.181	0.669	0.026	0.975	0.176	0.627	0.462	0.675	0.106	0.433
PCA 100 (with signaling gene space)	–	0.179	0.561	0.128	0.989	0.191	0.624	0.455	0.676	0.205	0.445
Dense	–	0.243	0.643	0.000	0.734	0.147	0.404	0.569	0.514	0.148	0.378
Dense with signaling pathways	–	0.259	0.648	0.000	0.849	0.236	0.486	0.509	0.656	0.130	0.419
Dense with PPI	–	0.196	0.638	0.041	0.927	0.212	0.550	0.575	0.686	0.179	0.445
Dense with PPI/GRN	–	0.194	0.645	0.007	0.930	0.294	0.542	0.600	0.711	0.190	0.457
Dense with PPI/GRN	100	0.068	0.771	0.182	0.956	0.849	0.561	0.415	0.553	0.710	0.563
Signaling pathways (+)	–	0.163	0.438	0.011	0.619	0.179	0.475	0.344	0.465	0.128	0.314
Signaling pathways (+)	100	0.149	0.307	0.050	0.402	0.198	0.352	0.592	0.368	0.137	0.284
Signaling pathways (parameter tuning) (+)	–	0.107	0.768	0.049	0.960	0.625	0.549	0.471	0.627	0.110	0.474
Signaling pathways (parameter tuning) (+)	Size*	0.155	0.803	0.117	0.955	0.568	0.550	0.497	0.623	0.150	0.491
scvis	–	0.203	0.522	0.000	0.813	0.077	0.733	0.424	0.683	0.512	0.441

*The size of second layer (size) is defined after tuning in hyperparameter tuning networks

ISSN: 1756-0381