In a recent publication in Medicinal Chemistry Letters, scientists from Insilico Medicine have called for stricter standards in the evaluation of molecules produced by generative artificial intelligence (AI). The team's perspective, part of the journal's special issue on "New Enabling Drug Discovery Technologies - Recent Progress," offers an in-depth analysis of AI and machine learning (ML) methods in modern drug discovery.

The researchers scrutinized eight molecular structures produced by generative chemistry over the past two years. They found that designing synthetically feasible molecular structures that are novel and experimentally valid remains a challenge for generative chemistry algorithms. 

"We aimed to provide a comprehensive analysis of the strengths of certain AI and ML generative chemistry approaches in producing truly novel and synthetically feasible molecular structures," said Alex Aliper, Ph.D., President of Insilico Medicine, who co-authored the study.

The team's focus extended beyond just AI-generated structures. They also examined the validity of these structures from the viewpoint of a medicinal chemist, including aspects of synthesis and biological assessment.

As terms like "generative AI" and "generative chemistry" become more commonplace, Insilico scientists believe it's crucial to better define these terms and validate the generated structures across various measures. Their recommendations include:

• A thorough inspection of generated structures for novelty and patentability.
• Use of balanced preprocessing rules and medicinal chemistry filters adapted for generative pipelines.
• Avoidance of misleading statements, such as "novel drug candidate" and "novel lead compounds," unless supported by comprehensive biological data.
• Use of stringent similarity metrics.
• Provision of all generated structures to medicinal chemists, not just those deemed most promising by authors.
• Evaluation of active AI-origin molecules using standard MTS or MTT assays to avoid nonspecific action and cytotoxicity.
• Assessment of synthetic accessibility.
• Improvement of the generative engine, with more attention to the training set, the test set, and similarity metrics.
• More focus on reinforcement learning with advanced systems and processes intended to rapidly evaluate the generated molecules for desired properties.

"We are encouraged by the increasing use of generative AI in chemistry, which can help speed and expand drug discovery efforts," said Alex Zhavoronkov, Ph.D., founder and CEO of Insilico Medicine and co-author of the paper. "However, we believe that publications in generative chemistry should always include experimental validation and rigorous evaluation and review by medicinal chemists."

‍