benchmarks/open_generation/MolOpt/readme.md

MolEdit

Target: Let the LLM edit the molecule and optimize the molecule with specific properties. Match if the generated molecule meets the standard of the requirements.

Subtasks

LogP

• Description: Optimize the molecule to have the higher or lower LogP value.
• Input: The instruction that specifies the target LogP value direction.
• Output: The optimized molecule SMILES
• Example:
  • Input: Please optimize the molecule CCO to have a higher LogP value.
  • Output: OCCO
• Evaluation Metrics:
  • Success Rate (MAIN): The percentage of generated molecules that meet the requirements.
  • Molecule Similarity: The edited molecule should make as few changes as possible to the original molecule.
  • Molecule Validity: The percentage of generated molecules that are valid.

MR

• Description: Optimize the molecule to have the higher or lower MR value.
• Input: The instruction that specifies the target MR value direction.
• Output: The optimized molecule SMILES
• Example:
  • Input: Please optimize the molecule CCO to have a higher MR value.
  • Output: CCCC
• Evaluation Metrics:
  • Success Rate (MAIN): The percentage of generated molecules that meet the requirements.
  • Molecule Similarity: The edited molecule should make as few changes as possible to the original molecule.
  • Molecule Validity: The percentage of generated molecules that are valid.

Toxity

• Description: Optimize the molecule to have the higher or lower toxicity.
• Input: The instruction that specifies the target toxicity direction.
• Output: The optimized molecule SMILES
• Example:
  • Input: Please optimize the molecule CCO to have a higher toxicity.
  • Output: CCN
• Evaluation Metrics:
  • Success Rate (MAIN): The percentage of generated molecules that meet the requirements.
  • Molecule Similarity: The edited molecule should make as few changes as possible to the original molecule.
  • Molecule Validity: The percentage of generated molecules that are valid.