Close to 50 years ago, researchers started working through ways to predict how proteins reach their three-dimensional structure. In 1990, the experiment CASP (Critical Assessment of Protein Structure Prediction) was created, confronting scientists to come up with a system to predict protein folding. Proteins are the most important element in any living organism. Understanding the shapes/structures of proteins means understanding the functions they have. This is particularly important for the field of medicine, especially in drug designing and treatments for cancer, Alzheimer’s, etc. Predicting protein structure can also be useful in many other fields.
DeepMind is one of the top artificial intelligence research companies in the world. It was founded by Demis Hassabis, Shane Legg and Mustafa Seluyman in 2010. They built the well-known model, “AlphaGo”, and many others using deep reinforcement learning. In 2016, DeepMind shifted its attention to the “protein folding problem” (aka Protein Structure Prediction). In 2018, DeepMind stood out with its artificial intelligence program AlphaFold. They managed to successfully predict 25 out of 43 proteins. This year, they got a prediction of about 90 GDT (Global Distance Test) with a model trained on approximately 170,000 samples . These results are almost at the same level of accuracy as the determining of protein shapes in a scientific lab.
AlphaFold is most likely going to reshape the way living things are being studied. The enormous costs of having high quality equipment in labs will be reduced. Time will not be spent on tedious lab tasks, instead, scientists will be focused on the questions that need to be asked and answered. To put it in the words of Andrei Lupas (evolutionary biologist at the Max Plank Institute of Biology in Tubingen, Germany); “This will change medicine. It will change research. It will change bioengineering. It will change everything”. AlphaFold has already helped Mr. Lupas find the structure of a protein.
I think what makes it unique is that AlphaFold made exponential progress since 2018. Additionally, the model performed really well considering the number of samples. Not many were expecting to have an AI as accurate as AlphaFold this soon. Such accurate results in solving the protein folding problem is a big leap forward for the science community and many other fields. This is a huge contribution for humankind on behalf of DeepMind. In the right hands and with the right intentions, AlphaFold will without a doubt propel us into a safer future.
Links: