Microsoft has unveiled a new deep-learning model called Biomolecular Emulator-1 (BioEmu-1). This model can “generate thousands of protein structures” hourly. The company has also released the model as open source. It is based on the preprint of a research study that was unveiled last December.
Microsoft said the model offers superior computational efficiency compared to traditional molecular dynamics (MD) simulations, thereby opening the door to insights that have been out of reach until now.
“Predicting a single protein structure from its amino acid sequence is like looking at a single frame of a movie – it offers only a snapshot of a highly flexible molecule,” Microsoft said. The BioEmu-1 provides researchers and scientists with a comprehensive view of all the different structures each protein can adopt.
“A deeper understanding of proteins enables us to design more effective drugs,” the tech giant added.
By training and fine-tuning extensive datasets that help the model recognise, map, predict, and sample protein structures, Microsoft said the model is capable of making accurate predictions for proteins it has never seen before.
“Our model offers a view on intermediate structures, which have never been experimentally observed, providing viable hypotheses about how this protein functions,” it further said.
BioEmu-1 accurately predicts MD equilibrium distributions using much less computational power. Microsoft compared 2D projections of the structural distribution of DE Shaw research’s simulation of Protein G and samples from BioEmu-1.
“BioEmu-1 reproduces the MD distribution accurately while requiring 10,000 to 1,00,000 times fewer GPU hours,” Microsoft said.
In December last year, Google DeepMind open sourced the AlphaFold 3 model, making its training weights accessible to academic researchers and scientists for non-commercial use.
Last year’s Nobel Prize in Chemistry was awarded to Demis Hassabis, CEO and co-founder of Google DeepMind, and John M Jumper for their contributions to protein structure prediction through AlphaFold. David Baker, a professor at the University of Washington, received the other half for computational protein design.
Microsoft is on a roll, and this announcement comes days after it unveiled the new Majorana 1 quantum chip. Satya Nadella, CEO of Microsoft, called Majorana 1 “a chip that can fit in the palm of your hand yet can solve problems that all computers on Earth today combined could not”.
The Majorana 1 uses a new ‘Topological Core’ architecture and can potentially hold one million qubits on a single chip, slightly larger than desktop computer CPUs.
