A novel artificial intelligence (AI) tool that deduces logical inferences regarding the role of unfamiliar proteins offers a promising avenue for scientists to unravel the intricacies of cellular functions.
Crafted by Maxat Kulmanov, a bioinformatics researcher at KAUST, along with his team, this tool surpasses existing analytical techniques in predicting protein functions and can even analyze proteins lacking clear matches in current datasets.
The study is featured in Nature Machine Intelligence.
Known as DeepGO-SE, the model leverages extensive language models akin to those utilized in generative AI tools like Chat-GPT. It then applies logical entailment to derive significant insights about molecular functions grounded in fundamental biological principles governing protein operations.
Essentially, it enables computers to logically deduce outcomes by constructing models of specific aspects of the world—such as protein function—and inferring the most plausible scenarios based on common sense and reasoning about the expected outcomes within these models.
“This approach holds numerous potential applications,” notes Robert Hoehndorf, the leader of the KAUST Bio-Ontology Research Group overseeing this project, “especially in scenarios requiring logical reasoning over data and hypotheses generated by neural networks or other machine learning models.”
Kulmanov and Hoehndorf, in collaboration with Stefan Arold from KAUST and researchers at the Swiss Institute of Bioinformatics, evaluated the model’s efficacy in deciphering the functions of proteins whose biological roles remain obscure.
By utilizing data on the amino acid sequence of a poorly characterized protein and its documented interactions with other proteins, the tool accurately predicted its molecular functions. DeepGO-SE exhibited such precision that it ranked among the top 20 out of over 1,600 algorithms in an international competition of function prediction tools.
The KAUST team is currently employing this tool to explore the functions of mysterious proteins identified in plants thriving in the harsh environment of the Saudi Arabian desert. Their objective is to uncover novel proteins with potential biotechnological applications and encourage fellow researchers to adopt this tool.
According to Kulmanov, “The capability of DeepGO-SE to analyze unexplored proteins can streamline various tasks including drug discovery, metabolic pathway analysis, disease associations, protein engineering, screening for specific proteins of interest, and more.”
Additional Details:
Title: Protein function prediction as approximate semantic entailment, Nature Machine Intelligence (2024). DOI: 10.1038/s42256-024-00795-w
Source: King Abdullah University of Science and Technology
Citation:
AI tool predicts function of unknown proteins (2024, February 14)
Retrieved on 14 February 2024
from https://phys.org/news/2024-02-ai-tool-function-unknown-proteins.html
This content is copyrighted. No part may be reproduced without written permission, except for fair use in private study or research. The information provided is solely for informational purposes.
