Skip to content

Improved Protein Structure Prediction Using Potentials from Deep Learning

“It will change medicine. It will change research. It will change bioengineering. It will change everything.”

Andrei Lupas, evolutionary biologist at the Max Planck Institute for Developmental Biology

As the first bundle of vaccines arrive to provide some relief to the global health scourge of COVID-19, another equally impressive scientific milestone this past week was perhaps less noticed. Google’s DeepMind and its AlphaFold algorithm was able to make a giant leap in using a protein’s amino acid sequence to predict a protein’s three dimensional structure (the “protein folding problem”).

The biennial Critical Assessment of Structure Prediction (CASP) challenge of protein prediction announced this month that DeepMind was the winner as it outperformed about 100 other teams (including solving the protein structure of Orf8, a coronavirus protein) at about 90% accuracy). This capability has very high level impact on biomedical science and drug discovery and obviates the need of specialized but tedious and expensive techniques such as cryo-electron microscopy.

This Nature paper, published earlier this year, elucidated the methodology that DeepMind deployed for this landmark achievement in molecular biology. The DeepMind team utilized a convolutional neural network and a gradient descent algorithm to make accurate predictions of the distances between pairs of residues that ultimately can help predict the structure of the protein.

The authors should be commended that they made the statements of data and code availability available. This accomplishment is perfect in its timing with the world in the grips of a global pandemic.

The article can be read here

The Edupreneur Business Model Canvas

One type of physician entrepreneur is an edupreneur. If we use the definition that entrepreneurship is the pursuit of opportunity with scarce or uncontrolled resources with the goal

Show Buttons
Hide Buttons