Skip to content

Latest commit

 

History

History
60 lines (32 loc) · 1.73 KB

README-Generation-Cond.md

File metadata and controls

60 lines (32 loc) · 1.73 KB

Conditional Generation with GraphGPT-C Decoder

The conditional generation of GraphGPT-C is similar to the unconditional version. However, there are some extra configurations to control the properties. For other configurations you can refer to Unconditional Generation.

Extra Generation Configurations

  • Conditions:
    • value_qed: (float) None (The target QED value for generated molecules. The model will not condition on this property if not specified.)
    • value_sa: (float) None (The target SA score for generated molecules. The model will not condition on this property if not specified.)
    • value_logp: (float) None (The target logP value for generated molecules. The model will not condition on this property if not specified.)
    • scaffold_smiles: (str) None (The target scaffold SMILES. The model will not condition on this property if not specified.)

Configurations in the Paper

We use the following configuration to test the ability of GraphGPT on conditioning molecular properties:

strict_generation="False"
fix_aromatic_bond="True"

do_sample="False"

check_first_node="True"
check_atom_valence="True"

Example scripts can be found in scripts/generation/unconditional/examples.

Generate with More Diversity

You can further turn on the probabilistic sampling for more diversity:

strict_generation="False"
fix_aromatic_bond="True"

do_sample="True"
top_k=4
temperature=1.0

check_first_node="True"
check_atom_valence="True"

Evaluate & Visualize the Results

Run scripts/generation/conditional/visualize.sh.

The mean and variance property values of generated molecules will also be saved to the summary.txt.