We’re poised for a new era of medical advances. A key step: understanding the molecular machines that govern health and disease
Biotech Voices is a collection of exclusive opinion editorials from some of the leading voices in biopharma on the biggest industry questions today. Think you have a voice that should be heard? Reach out to senior editors Kyle Blankenship and Amber Tong.
We witnessed another important advance in science last year, when London-based researchers at DeepMind announced a major leap forward in the accuracy of an artificial intelligence (AI) platform called AlphaFold, which is designed to rapidly and reliably predict the three-dimensional structure of proteins. It is another breakthrough in our understanding of proteins in the last 70 years, and one that could help usher in a new era of medicine.
Looking back to the start of this revolution, elucidation of the double helix structure of DNA in 1953 began our quest to decipher the blueprint of life. The initial draft sequence of the human genome published in 2000 made available the instruction manual for the functional embodiment of genes: proteins.
In recent decades, the three-dimensional architecture encoded by a protein’s amino acid sequence has become increasingly accessible thanks to advances in x-ray-based images originally used to reveal the structure of the double helix. Formidable new tools like cryogenic electron microscopy (cryo-EM) allow us to develop a sophisticated understanding not just of the form of folded proteins, but a better understanding of the “pocketome” at a level of detail that can explain protein function and dysfunction in disease.
This isn’t just an academic exercise. Determining a protein’s structure can provide vital clues to accelerate the design of new medicines. This type of insight has been of global interest in 2020. The amazing progress the biopharma industry has made in developing Covid-19 vaccines and therapeutics was possible, in part, because scientists identified the sequence and structure of the SARS-CoV-2 virus’s spike protein in record time.
The ability of scientists to accurately map protein structures — let’s call them the Google Maps of proteins — could provide critical information about their function and dysfunction and inform medicine discovery. Researchers around the world have been making steady progress in obtaining these maps; the number of publicly available protein structures quadrupled in the last 20 years. However, the majority of proteins encoded by the human genome have not been solved and obtaining an accurate structure can take years of lab time.
DeepMind’s AlphaFold, which was recently honored as a runner-up in Science’s prestigious Breakthrough of the Year awards, is a major advance set to be a major breakthrough in protein structure prediction. It deploys an AI algorithm, trained on hundreds of known protein folds and structures, to predict unsolved protein structures based solely on the amino acid sequence. AlphaFold’s incredible performance in replicating some known protein structures was demonstrated in a recent international competition. The implication is that AIphaFold could complete in days or weeks structure determination that would typically take years for many proteins.
Exciting as this news is, there are a multitude of complex steps between solving a three-dimensional structure, determining disease mechanisms, and discovering safe and effective drugs.
When proteins are at work they are not static in the way they often appear in scientific journals, like a sports car parked in the showroom of a dealership. They are dynamic, high-performance machines, whose exquisite engineering is most apparent when they are in motion. Medicinal chemists are at a great advantage when they have access to the “map” of a protein and can visualize the bends, twists and folds that occur as proteins transition from an inactive to active state — also known as protein motion. The level of detail in protein “maps” is critical when navigating the challenges of drug discovery. It will be interesting to see whether AlphaFold provides the crucial level of resolution chemists need.
The efficient design of drugs that bind to relevant orthosteric or allosteric pockets in proteins to modify function for therapeutic benefit requires resolution at the level of individual atoms, equivalent to seeing individual houses in a Google map. Drug discovery scientists optimize initial molecules to achieve the optimal balance across a broad range of critical properties in order to turn them into viable clinical candidates. If decoding a protein structure is challenging, discovering promising therapeutics is a nearly Herculean task. Every one of these steps is laborious and expensive. It’s no wonder that it can take many years, and cost many millions, just to get to the point of having a molecule selected for preclinical studies.
Fortunately, advances in computing power and computational chemistry techniques have brought us ever-more-sophisticated design and binding simulation tools. When coupled with detailed protein structures and molecular dynamics information, these computational methods enable medicinal chemists to conduct key design and optimization steps “in silico.”
For example, accurate physics-based models can help chemists prioritize promising leads from billions of the approximately 1050 potential drug-like molecules that exist in the vastness of chemical space before taking the costly and time-consuming step of synthesizing and testing them in the lab. This less empirical approach, fueled by access to high-resolution protein structures, will ultimately allow drug discoverers to design higher quality drugs as well as tool molecules for use in validation of targets to help prioritize therapeutic hypotheses.
My opening statement asserted that we’re at the dawn of a new era in medicine. We’ve reached this point thanks to a combination of unprecedented computing power, advances in molecular simulation technologies and new insights into human biology. We can begin to integrate our knowledge of the genome, the proteome, the interactome, and the pocketome to design drugs with increased speed and accuracy. With tools such as AlphaFold to speed insights into the molecular machines that govern health and disease, coupled with computational modeling to aid drug design, the next decades are sure to yield many more medicines.
I liken these connected advances to taking the blindfolds off drug hunters. And all of us stand to benefit.
Karen Akinsanya is executive vice president, chief biomedical scientist, and head of discovery R&D at Schrödinger, Inc.