OpenAI's GPT-4 Can Simulate Simple Amino Acid and Protein Structures with High Precision
TapTechNews August 22nd news, the tech media TheDecoder released a blog post yesterday (August 21st) reporting that a study from Rutgers University shows that the GPT-4 language model of OpenAI company can simulate simple amino acid and protein structures with high precision.
The related research results were published in 'Scientific Reports'. The research team used the GPT-4 AI language model to explore its performance in basic structural biology tasks and found that the AI model can accurately predict molecular structures.
Scientists asked GPT-4 to build three-dimensional structure models of 20 standard amino acids, and accurately predicted the atomic composition, bond lengths and angles in the feedback results, but GPT-4 made mistakes when simulating cyclic structures and stereochemical configurations.
In another experiment, GPT-4 was asked to simulate the common protein structure element - the structure of the α-helix, which requires integrating the Wolfram plug-in for mathematical calculation, and the result was comparable to the experimentally determined α-helix structure.
Also, GPT-4 also analyzed the binding between the antiviral drug Nirmatrelvir and the main protease of SARS-CoV-2. The model correctly identified the amino acids involved in the binding and accurately specified the distances between the interacting atoms.
Since GPT-4 is not specifically developed for structural biology tasks, these capabilities are very outstanding. The researchers pointed out that the modeling method of GPT-4 is not yet clear. It can use the existing atomic coordinates in the training data set or recalculate the structure from scratch - further extensive research is needed to draw a definite conclusion.
The researchers said that dedicated AI tools such as AlphaFold3 can predict more complex structures, while GPT-4 is expected to complete basic structural biology tasks. This modeling ability is still very primary and has limited practical applications.
Nonetheless, the research team said that this study sets a precedent for applying this technology in structural biology. The researchers suggest further studying the capabilities and limitations of generative AI and can further explore the application of AI in other potential life science fields beyond the field of structural biology.
TapTechNews attaches the reference address