Extended Data Table 2 Comparison of performance between AMPGenix and other currently available unconstrained generative models

From: A generative artificial intelligence approach for the discovery of antimicrobial peptides against multidrug-resistant bacteria

Model

Uniqueness

Diversity

Novelty

FCD

Diversity*

Novelty*

FCD*

AMPGenix-T0.5

0.81

0.97

0.99

13.61

0.90

0.98

14.41

AMPGenix-T1

0.97

0.98

0.99

10.21

0.94

0.98

14.98

AMPGenix-T2

0.99

0.98

0.99

9.26

/

/

/

AMPGenix-T3

1.00

0.99

0.99

9.57

0.93

0.98

16.87

ProteoGPT-T0.5

0.91

0.87

1.00

23.65

   

ProteoGPT-T1

0.90

0.97

1.00

14.87

   

ProteoGPT-T2

1.00

0.98

0.99

13.07

   

ProteoGPT-T3

1.00

0.98

0.99

13.01

   

HydrAMP

 

0.96

0.99

22.82

   

Basic

 

0.99

0.99

14.40

   

PepCVAE

 

0.99

0.99

13.45

   

AMP-GAN

 

0.98

0.99

11.54

   

AMP-LM

 

0.99

0.99

10.91

   
  1. Note: Diversity, Novelty, and FCD were calculated using only the generated sequences containing standard AAs, compared with the known AMPs containing only natural AAs (n=15,463); Diversity*, Novelty*, and FCD* were calculated using the generated sequences containing unnatural AAs, compared with all known AMPs (n=16,062).
Back to article page