Key points from article :
AI is being explored to improve drug discovery by analyzing molecular structures and predicting drug interactions. It can identify potential drug targets, optimize molecules, and assess unwanted side effects. However, AI’s effectiveness depends on high-quality biological data, which often lacks consistency due to variations in experimental methods.
Researchers emphasize the need for standardized data collection to enhance AI predictions. Organizations like the Human Cell Atlas provide structured biological data for better AI training. However, many existing public databases still contain inconsistencies, limiting AI’s ability to make accurate comparisons.
The lack of negative results in scientific publications creates bias in AI models. Many failed experiments go unpublished, leading AI to favor positive results. This can misguide AI-driven drug discovery by overrepresenting successful compounds while ignoring critical failures that might offer valuable insights.
Pharmaceutical companies possess vast, well-organized datasets but hesitate to share them due to competition. Some projects, like Melloddy, enable secure data-sharing among companies without exposing proprietary information. However, these efforts do not improve public datasets that academic researchers rely on for drug development.
AI-driven companies like Insilico Medicine link government research data to clinical trials and patents to enhance drug discovery. Their AI platform identified a promising drug candidate for fibrotic diseases, which has completed phase IIa trials. “Since 2019 we’ve nominated 22 preclinical candidates,” says Alex Zhavoronkov, founder and chief executive of Insilico Medicine.
Despite challenges, AI is making progress in drug discovery by generating and testing drug candidates faster. However, improvements in data sharing, standardization, and inclusion of negative results are crucial for AI to reach its full potential in medicine. The research was covered by Nature Outlook: Robotics and Artificial Intelligence.
The research was covered by Nature Outlook: Robotics and Artificial Intelligence.
