Data1 | Genes | Cells | Ref2 | Estimation methods3 |
|---|
ZIGeneNet | GeneNet | scLink | Pearson |
|---|
[50] | 570 | 108 | TF | 1/83 | 1/83 | 0/13 | 1/151 |
Protein | 728/6407 | 728/6407 | 28/935 | 347/4552 |
All | 11.4% | 11.4% | 3.0% | 7.6% |
[51] | 2608 | 127 | TF | 1/116 | 1/109 | 40/11435 | 5599/2006048 |
Protein | 24/116 | 22/109 | 437/11435 | 118793/2006048 |
All | 20.7% | 20.2% | 4.2% | 6.2% |
[52] | 1010 | 1064 | TF | NA | NA | NA | NA |
Protein | 48/321 | 52/365 | 30963/203718 | 19764/135341 |
All | 15.0% | 14.25% | 15.2% | 14.6% |
[53] | 3860 | 8715 | TF | 0/197 | 0/234 | 56708/1422895 | 102516/3933970 |
Protein | 21/197 | 22/234 | 4018/1422895 | 18937/3933970 |
All | 10.7% | 9.4% | 4.2% | 3.1% |
- Results are shown as positive predictive value, percentage or fraction between database-matching interactions and total estimated edges. Results of experimental data analysis are ordered based on size of count data. In analysis of [50, 51, 53] data, Stein-type shrinkage including ZIGeneNet and GeneNet have high precision scLink (Lasso-type shrinkage) and Pearson correlation when comparing to existing interaction databases. On the other hand, all methods perform similarly in [52] analysis
- 1Experimental scRNAseq data from Schizosaccharomyces pombe [50], Saccharomyces cerevisiae [51], Plasmodium falciparum [52] and Mus musculus [53]. Each dataset has corresponding number of selected genes and cells
- 2Reference database (All) including transcription factor (TF) database for gene regulatory interactions, protein-protein (Protein) interaction database. TF database is unavailable for Plasmodium falciparum in which results are shown as not applicable (NA)
- 3ZIGeneNet for GeneNet shrinkage using zero-inflated negative binomial modelling, GeneNet for GeneNet method without modification, scLink [49], Pearson for Pearson correlation
