Skip to main content

Table 4 Comparison of properties for structures generated by the canonical Prior, the reduced Prior, and corresponding Agents

From: Molecular de-novo design through deep reinforcement learning

Model

Prior

Agent

Priora

Agenta

Fraction valid SMILES

0.94

0.99

0.94

0.99

Fraction predicted actives

0.03

0.97

0.02

0.96

Fraction similar to train active

0.02

0.79

0.02

0.75

Fraction similar to test active

0.01

0.46

0.01

0.38

Fraction of test actives recovered (\(\times 10^{-3}\))

13.5

126

2.85

72.6

Probability of generating a test set active (\(\times 10^{-3}\))

0.17

40.2

0.05

15.0

  1. aDRD2 actives witheld from the training of the Prior