Table 4 Results for the zero-shot image-to-image mechanism of action (MoA) classification task

From: CLOOME: contrastive learning unlocks bioimaging databases for queries with chemical structures

Split

Method

Accuracy [%]

  

Top-1

95% CI

Top-5

95% CI

Top-10

95% CI

Random

CLOOME

13.0

[12.3, 13.7]

35.3

[34.3, 36.3]

48.0

[46.9, 49.0]

 

GapNet

1.09

[0.882, 1.33]

4.68

[4.25, 5.14]

8.63

[8.06, 9.24]

 

CellProfiler

2.87

[2.53, 3.24]

9.46

[8.86, 10.1]

14.9

[14.2, 15.7]

Scaffold

CLOOME

18.5

[17.3, 19.7]

47.9

[46.4, 49.5]

61.3

[59.7, 62.8]

 

GapNet

1.92

[1.52, 2.39]

9.00

[8.14, 9.92]

17.2

[16.1, 18.4]

 

CellProfiler

4.54

[3.92, 5.22]

15.3

[14.2, 16.4]

24.5

[23.2, 25.8]

  1. Given a molecule-perturbed microscopy image, the image corresponding to the matched molecule must be selected from a set of candidates. Top-1, top-5, and top-10 accuracy in percentage are shown for a hold-out test set, along with the upper and lower limits for a 95% confidence interval (CI) (n = 8826 for the random split and n = 4056 for the scaffold split) on the resulting proportion. The best method in each category is marked in bold.
Back to article page