Skip to main content

Table 3 Test comparison between the KLD–RF model and DTI prediction models

From: Random-forest model for drug–target interaction prediction via Kullback–Leibler divergence

Model

Test Set

Drug structure (Sim)

Similarity metric

Highest recall

AUC

Refs.

KLD-RF

17 Targets in ChEMBL

Multiple Conformers (3D-Sim)

KLD vector from TC

1.00

Average:

0.889

Average:

0.992

HSP90:

0.998

This Work

CSNAP3D

6 Targets in DUD

One Conformer with Lowest Energy

(3D-Sim)

28 including TC with cut-off 0.85

0.98

AUC*

0.54—0.70

HSP90:

0.79

Lo et al. [25]

CSNAP2D**

6 Targets in DUD

2D Structure

(2D-Sim)

TC with cut-off 0.6

0.83

Lo et al. [24]

SEA**

TC with cut-off 0.57

0.64

0.972***

Keiser et al. [47]

PASS**

Probability Function

0.11

Lagunin et al. [56]

SwissTarget

17 Targets in ChEMBL

2D + 3D-Sim

Probability Function from 2 and 3D TC

0.99

Average:

0.748

Average:

0.869

Gfeller et al. [58]

  1. AUC of CSNAP3D*: the average area-under-curve (AUC) was calculated from the curve having rank orders (%) as x-axis and TPR as y-axis. The AUC range was achieved from used different Sim metric. CSNAP2D, SEA, and PASS**: the described performance metric, TPR and AUC were citied from CSNAP3D [25]. The AUC of SEA was citied [59]