DeepMind, a British artificial intelligence company, recently claimed that they solved one of biology’s greatest problems, but their announcement has received a mixed response in the scientific community.
On 30 November, DeepMind announced that they had solved a 50-year-old problem in biological sciences – they could successfully predict how proteins fold. This announcement attracted a media frenzy and much excitement among certain scientists.
“This will change medicine,” said Andrei Lupas, an evolutionary biologist at the Max Planck Institute for Development Biology in Germany.
However, other scientists have voices scepticism over the true value of the research.
The Problem
Proteins are the building blocks of life. Almost every single bodily function involves proteins during some step of the way. The exact structure of different proteins is determined by our DNA.
The precise order of bases in DNA is split into groups of 3 – each group of 3 codes a different ‘amino acid’ which form long chains to create proteins.
The specific functionality of each protein is caused by how they fold together. The forces exerted by the atoms that make them up exert different forces on each other. This causes them to bend and fold into very particular shapes. These shapes then play a very important role in their usage.
For example, the protein, haemoglobin is found in blood. Its role is to take oxygen from the lungs to different parts of the body. The shape features a prominent ‘dimple’ that allows oxygen molecules to feat neatly inside and get carried to where it needs to go.
Many medical conditions, like Alzheimer’s and Parkinson’s are caused by proteins not folding properly and developing treatments for them hinges on understanding these folding ‘errors’.
The aim, in the last 50 years, has been to understand exactly how proteins fold up and we may finally have an answer.
The Software
Deepmind’s software, known as AlphaFold, was entered into the CASP (Critical Assessment of Structure Prediction). CASP is a structure prediction challenge, where different protein prediction pieces of software go up against each other.
AlphaFold went up against 100 other teams and came out on top. Nearly two thirds of its preidctions were comparable to the quality in experimentally determined structures.
Usually, when scientists attempt to work out how proteins have folded from a sequence of amino acids, they experimentally characterise is using a process known as X-ray diffraction. However, X-ray diffraction results are usually complex and take in-depth analysis to interpret. Because of this, it can take many years to get a full picture.
For example, Andrei Lupas had spent a decade trying to understand the shape of a particular bacterial protein to no avail, yet AlphaFold figured it out in 30 minutes.
This goes to show how much AlphaFold could change the entire field of structural biology.
The Response
Whilst some scientists responded with excitement, others have held back. Max Little, associate professor and senior lecturer in computer science at the University of Birmingham, urged caution, citing that it had only shown effectiveness within the context of the challenge.
“We can’t really be sure how well AlphaFold will work when faced with the far more rich and varied array of proteins found in the real world of living organisms” he told Business Insider. He added that it had shown potential only ‘within the context of the CASP database challenge’.
Professor Michael Thompson, structural biology professor at the University of California echoed these sentiments and questions how much use this will have practically. He wrote on Twitter that DeepMind could ‘never live up to the promise that’s been made’.
The true impact of this code will become clearer in time. There is no denying, however, that it is a big step forward to curing diseases once thought incurable.