The Need for Physics-Based Verification
AI Protein Models Generate Impossible Structures
A new RPI study finds AlphaFold2 and RoseTTAFold2 often produce chemically impossible results.
A digital visualization of a protein structure showing errors or impossible geometric connections, representing AI prediction failures.
Photo: Avantgarde News
Researchers at Rensselaer Polytechnic Institute (RPI) found that leading AI models frequently create impossible protein structures [1][2]. The study, published in PNAS, analyzed tools like AlphaFold2 and RoseTTAFold2 [1]. These models sometimes generate results that defy physical and chemical laws [1][2].
The findings highlight "growing pains" for AI in scientific laboratories [1]. Researchers emphasize that AI-driven biological research still requires physics-based verification to ensure accuracy [1][2]. Without these checks, scientists risk using flawed data in drug discovery and molecular biology [1].
Editorial notes
Transparency note
AI assisted drafting. Human edited and reviewed.
- AI assisted
- Yes
- Human review
- Yes
- Last updated
Risk assessment
Only two independent source domains were available in the input, which is below the recommended threshold of three.
Sources
Related stories
View all
Topics
About the author
Avantgarde News Desk covers the need for physics-based verification and editorial analysis for Avantgarde News.