Google’s machine learning can help discover new antibodies, enzymes and foods
DeepMind, a subsidiary of Alphabet (the parent company of Google), has shown that machine learning (ML) can predict the shape of protein machinery with unprecedented accuracy, paving the way for researchers to discover new antibodies, enzymes and foods.
The shape of a protein provides very strong clues as to how the protein machinery may be used, but does not fully resolve this question.
“So we asked ourselves: can we predict what function a protein performs? said Max Bileschi, staff software engineer, Google Research, Brain Team.
In an article in Nature Biotechnology, Google described how neural networks can reliably reveal the function of “dark matter” in the protein universe, surpassing state-of-the-art methods.
DeepMind worked closely with internationally recognized experts from EMBL’s European Bioinformatics Institute (EMBL-EBI) to annotate an additional 6.8 million protein regions in the “Pfam v34.0 database” version, a repository worldwide for protein families and their function.
These annotations exceed the expansion of the database over the past decade and will enable the world’s 2.5 million life science researchers to discover new antibodies, enzymes, foods and therapeutics.
For about a third of all proteins produced by all organisms, we simply don’t know what they do.
“It’s kind of like we’re in a factory where everything is buzzing, and we’re surrounded by all these awesome tools, but we only have a vague idea of what’s going on. Understanding how these tools work and how we can use them, which is where we think machine learning can make a big difference,” said Lucy Colwell, senior researcher at Google Research, Brain Team.
The Pfam database is an extensive collection of protein families and their sequences.
“Our ML models helped annotate an additional 6.8 million protein regions in the database,” the researchers said.
The company also launched an interactive science paper where “you can play with our ML models – get real-time results, all in your web browser, with no configuration required.”
According to the researchers, combining deep models with existing methods significantly improves remote homology detection, suggesting that deep models learn complementary information.
This approach extends Pfam coverage by more than 9.5%, surpassing additions made over the past decade, and predicts the function of 360 human reference proteomic proteins without prior Pfam annotation.
“The results suggest that deep learning models will be a central part of future protein annotation tools.”
(Only the title and image of this report may have been edited by Business Standard staff; the rest of the content is auto-generated from a syndicated feed.)