Yesterday was a good news day. In the 20 minutes that I had the radio on there were two really positive stories.
The story about Pfizer's Covid-19 vaccine gives real hope that we might be able to return to something resembling normality after this strangest of years. The speed with which their vaccine and other similar ones have been brought to this stage is unprecedented. No one reading this needs reminding that Covid-19 has blighted everyone's lives and presented new and unwelcome challenges to those already struggling with Parkinsons so this is a breakthrough worth shouting about.
I was still smiling about that story when Nick Robinson of BBC News started talking to Demis Hassabis of DeepMind about how they have largely solved the 50 year old problem of protein folding. Despite the enormous significance of the vaccine, I don't think that it's an exaggeration to say that this could be just as important.
While current treatments for Parkinsons help a lot with managing symptoms, it's fair to say there is room for improvement so drug design research is of particular interest. A big part of the science of drug design is analysis of how different molecules interact with the proteins that are responsible for the condition being studied. Determining the precise 3D structure of those proteins is fundamental to the process and can be found using traditional lab techniques such as
X-ray crystallography and
NMR spectroscopy. These are reliable but time consuming. There have been huge strides in genome sequencing over the last 30 years as techniques have been improved and, as a consequence, it has become cheaper. Translating DNA sequences to protein sequences is relatively trivial. However using that protein sequence to predict protein structure is fiendishly difficult since there are infinite permutations of how the proteins' amino acids can be folded. This conundrum has been a major bottleneck in the in the field of drug discovery among others. Every two years since 1994 CASP (Critical Assessment of protein Structure Prediction) has run a challenge that compares software generated 3D structures against the reliable but slower lab techniques. DeepMind's
AlphaFold is the first to achieve comparable results. AlphaFold uses artificial intelligence techniques to translate amino acid sequences into protein structures using the accumulated knowledge stored in a long established international
protein database to inform it. The AI learns fast so that structures can be predicted in days rather than years.
This breakthrough should rapidly accelerate drug discovery and lead to better targeted treatment of conditions such as Parkinsons. Unfortunately, the necessary regulatory pathway for bringing a drug to market means that, even with accelerated discovery, a viable new treatment will still be years away. Nevertheless, all sources of hope are very welcome and I look forward to hearing how this and other avenues develop.
Comments
Post a Comment