AI has made significant strides in decoding proteins, with positive implications for drug delivery and design
While most approaches to decode structures tend to be experimental, leaps in artificial intelligence has helped speed up the process.
Google’s DeepMind algorithm’s unprecedented accuracy in decoding of protein structures—this was recently reported in Nature—will help researchers develop better medication in the future. Although scientists have been studying protein structures for long now, the complexity of some molecules and the sheer combinations involved have led to the decoding of only 170,000 proteins, from over 200 million known to humans. Understanding of the structure of complex proteins still eludes mankind. While most approaches to decode structures tend to be experimental, leaps in artificial intelligence has helped speed up the process. In 2018, when DeepMind first participated in the bi-annual competition to determine new protein structures, it scored 15% higher than everyone else and achieving a GDT—a 0-100 scale that determines the accuracy of prediction—score of 60. Other approaches, until then, had only been able to achieve scores close to 40.
However, this year, the algorithm, AlphaFold, was able to achieve a score of 92.4 for less complex structures and 87 for complex molecules. Also, given that each team has to share data on how it arrives at the calculation, this will also help other researchers tweak other techniques for better efficacy. If drug designers can isolate every protein molecule and understand its structure, it will also help deliver more drugs to the market. While AlphaFold wasn’t able to solve some protein structures, with time and training, it will be able to do that as well. For now, it is a significant achievement towards understanding drug responses.