Google DeepMind uses Antigravity AI to identify potential new antimicrobial peptide
Researchers found a possible new antimicrobial peptide in a single AI-assisted session. Laboratory tests are now underway to see whether it can act against germs linked to respiratory infections.
Google DeepMind researchers used Antigravity to find a possible new antimicrobial peptide. Laboratory tests are now checking whether it works as predicted.
Google DeepMind researchers have used Google Antigravity, the company's AI-powered development platform, to find a possible new antimicrobial peptide that had not been studied before. The work shows how AI agents can help scientists make discoveries by bringing different research tools together in one process.
The possible new peptide is predicted to act against germs linked to respiratory tract infections. Researchers found it during an interactive session. A researcher guided the investigation, while Antigravity searched databases, analyzed proteins and reviewed existing scientific studies.
The discovery has so far been made using computers. The peptide and several other possible candidates found through the same process are now being tested in a laboratory to see whether the predictions are correct.
Google DeepMind has been expanding Antigravity beyond its original use in coding. It now combines its Gemini AI models with specialist scientific tools, including AlphaFold and AlphaGenome.
Pushmeet Kohli, Chief Scientist at Google Cloud and VP Science at Google DeepMind, described how quickly the discovery happened in a LinkedIn post:
"As an example, in a single session using Antigravity, our scientists working on Antimicrobial resistance surfaced a novel antimicrobial peptide candidate."
How Antigravity searched millions of proteins to find a possible candidate
The research focused on antimicrobial peptides. These are molecules found in nature that can act against bacteria. Scientists are studying them as possible alternatives to traditional antibiotics.
Finding useful candidates means searching huge biological databases. Many of these databases contain protein sequences whose functions scientists have not yet identified.
Google DeepMind used ProtNLM, its Protein Natural Language Model, to help with this task. The model studies amino acid sequences, which are the building blocks of proteins, and predicts what those proteins might do. This gives researchers clues about proteins that have not been studied before.
In the demonstration, researchers asked Antigravity to look for possible antimicrobial peptides in Streptococcus pseudopneumoniae.
The system reduced millions of database entries to a small group of possible candidates. It found a peptide made up of 97 amino acids that belongs to the Lactococcin 972 family.
The researchers then compared the candidate with reviewed protein sequences, examined its predicted three-dimensional shape using AlphaFold data and studied nearby genes for other biological clues that might support the findings.
The analysis found no meaningful match in the amino acid sequence among proteins that had been manually reviewed. However, comparisons of the predicted structure showed similarities to a bacteriocin whose structure had been established through experiments. A bacteriocin is a type of antimicrobial peptide made by bacteria.
Antigravity also searched existing scientific studies to develop ideas about how the molecule might work and which germs it might affect.
These results provide computer-based evidence that the candidate is worth studying further. They do not prove that it works as an antimicrobial treatment.
A new 3D viewer and research credits add to the scientific tools
Alongside this example, Google DeepMind has introduced a Biomolecular Structure Viewer in Antigravity. It lets researchers and AI agents look at and interact with three-dimensional protein structures in the same workspace.
The viewer can display structures from the AlphaFold Protein Structure Database or predictions made by researchers themselves. Scientists can highlight parts of a structure, add notes and ask the AI agent questions about specific features of a molecule. The researcher and AI system can both see these interactions.
The company has also introduced an Antigravity for Science credits program to help academic researchers use the tools. However, the announcement does not say how much the credits are worth or give detailed rules about who can qualify.
The next step in the antimicrobial peptide project is to test the predictions through experiments. Researchers are testing the identified candidate and other candidates found through the same process to see whether their predicted biological activity can be confirmed in the laboratory.