Table 1 Comparison of \(75^{th}\) percentile of gene pairs intersected with the top 400 gene pairs (ground truth).

From: Knowledge graph-aided Bayesian active learning for top-K genetic interaction discovery

Gene pair

K=20

K=1

Difference

CNOT1-RRM2

0.861

0.250

0.611

CNOT1-TRIM55

0.806

0.222

0.583

EIF3M-RTF1

0.944

0.389

0.556

EIF3D-RTF1

0.833

0.278

0.556

RTF1-TRIM55

0.833

0.306

0.528

EIF3A-SUPT5H

0.833

0.417

0.417

PTGES3-RTF1

0.972

0.639

0.333

EIF3F-RTF1

0.889

0.556

0.333

EIF3E-RPL3

0.944

0.639

0.306

EIF3C-SUPT5H

0.889

0.583

0.306

CTR9-EIF3M

1.000

0.722

0.278

CTR9-RPL3

0.972

0.694

0.278

EIF3E-RRM2

0.972

0.694

0.278

NUP98-TNPO3

0.944

0.667

0.278

RBM17-TNPO3

0.972

0.722

0.250

CPSF6-SUPT5H

0.944

0.722

0.222

RRM2-SUPT5H

0.889

0.667

0.222

CUL5-TNPO3

0.806

0.583

0.222

DARS-TNPO3

0.889

0.722

0.167

CCNK-RUVBL1

0.778

0.611

0.167

RANBP1-RANBP2

0.778

0.611

0.167

RUVBL1-SRRM2

0.833

0.694

0.139

PPIA-TNPO3

0.806

0.667

0.139

CCNT1-RUVBL1

0.778

0.667

0.111

AHCYL2-TNPO3

0.778

0.694

0.083

COPS2-VIRMA

0.778

0.694

0.083

EIF3H-RUVBL1

0.778

0.722

0.056

  1. This list shows those gene pairs chosen by the \(K=20\) model at least \(75\%\) of the time in our numerical experiments, but that were chosen less than \(75\%\) of the time by the \(K=1\) model. Note that there were no gene pairs chosen by the \(K=1\) model more than \(75\%\) of the time and by the \(K=20\) model less than \(75\%\) of the time. In this way the list represents those gene pairs that account for the additional performance gain by the more complex model. The gene pairs are sorted by the difference in percentile so that gene pairs toward the top are more discriminative of the two models.
Back to article page