"AI Drug Design Models Fall Short in Protein Interaction Physics"

A study from the University of Basel reveals that state-of-the-art AI programs, including those predicting Protein structure interactions, fall short by merely memorizing patterns rather than grasping underlying physics.

• Researchers at the University of Basel published findings in Nature Communications, indicating that Artificial intelligence models struggle to predict Protein interactions with new Ligand molecules critical for drug development.
• The study reveals that these AI programs do not understand the physical relationships governing Protein structure but instead rely on memorized patterns, limiting their effectiveness in innovative drug discovery.
• This research highlights the limitations of current AI in drug design, particularly in predicting interactions involving Active ingredient, enzymes, and antibodies, which are essential for developing targeted therapies.