May 23, 2023
Mizzou engineers received funding from the National Science Foundation to develop a tool to predict protein function based on amino acid order.
Jianlin “Jack” Cheng envisions developing open source software that allows users to enter sequences. The system would then be able to predict not only how strings of amino acids form structures, but also the roles they play in cells. Moreover, the system pinpoints specific sites in proteins that perform functions.
Because proteins are the building blocks of life, their applications range from genetic engineering of drought-tolerant crops to advanced drug development.
“This will allow researchers to understand what molecular function the protein has,” said Cheng, a Thompson professor of electrical engineering and computer science. “For example, if a protein drives tumor growth in cancer patients, scientists may be able to design drugs that block its active site to slow or stop growth.”
Cheng uses a deep transformer model. This is a large-scale language model that bears some similarities to the model that powers his ChatGPT, a popular generative artificial intelligence (AI) program that generates text based on user prompts. Like words, protein sequences are the language of biological systems.
The team is developing three deep trance models. One-dimensional array-based transformers consider sequences of amino acids. The 2D Graph Transformer explores how proteins interact and analyzes how these interactions affect them. The 3D Equivalent Graph Transformer also takes into account protein structure and the different sites within the protein that perform specific tasks.
This is the latest milestone in Cheng’s impressive career in protein prediction. In 2012, he and his students first demonstrated the superiority of deep learning for predicting protein structures in 10 fields.th Critical Evaluation of Protein Structure Prediction Techniques (CASP10). At the CASP14 experiment in 2020, Google-owned Deep Mind presented AlphaFold2, an advanced deep learning method that predicts protein structure with unprecedented accuracy. In the 2022 CASP15 experiment, the Cheng Group further improved the accuracy of AlphaFold2-based protein structure prediction by 8-10%.
“We are leveraging state-of-the-art protein structure methods and using AlphaFold2 for this particular project,” Cheng said. “Language modeling methodologies are fairly new to the field. It’s an interesting area and we’re putting a lot of research effort into it. We’re very excited about this work.”
Read more about Cheng’s work in the field of protein prediction.
