Google frequently dominates headlines across a variety of sectors and industries— ranging from the advertising business and intelligent search to self-driving technology and personal devices. However, one additional interest that is perhaps less well-known is the company’s significant investments in the world of biology and drug discovery.
Among the most important aspects of this story is AlphaFold, the artificial intelligence and foundation model platform which has enabled DeepMind and Google to make incredible strides in the world of biology. Specifically, its developers sought to help alleviate key challenges that researchers have in the scientific process; one such issue of paramount importance is the ability to determine the structure and sequence of proteins— the building blocks of living organisms. Although all 300,000,000 proteins on Earth are just a combination of ~20 basic amino acids, the sequence and folding of these elements are the keys to life’s essential functions. Therefore, the inspiration and creation of AlphaFold was to boldly unravel the mystery of protein folding, and thereby, unlock new potential breakthroughs in science.
To do so, scientists developed AlphaFold by training it with nearly 100,000 known proteins. Two major iterations later and in partnership with Isomorphic Labs, AlphaFold 3 is the latest model that has been released, and boasts significant accuracy in understanding protein interactions and modeling capabilities.
Dr. John Jumper, PhD, is one of the key pioneers behind the development of AlphaFold, and enthusiastically describes the “AlphaFold story” as a significant milestone in science, especially with regards to how protein structure and folding is understood: what previously took years can now be completed in mere minutes with the platform.
Why is this important for the masses?
Because this technology has numerous potential applications. Perhaps one of the most important is its immense boon to the pharmaceutical and drug discovery industries. A drug/medication is a essentially just a small molecule that ultimately attaches to a protein inside the body in some configuration to trigger a cascade of events to target a very specific pathology in the body. AlphaFold and DeepMind’s work in this arena have unlocked unprecedented capabilities in targeting these proteins for the sake of drug discovery and development. As scientists better understand protein structures and interactions, they can also create better targets for medications, further understand side effects and venture into new arenas of protein-drug interactions that may not have been previously fathomable.
As Max Jaderberg, Chief AI Officer at Isomorphic Labs describes, the work marks a monumental chapter in human history, as scientists have been enabled “to rationally develop therapeutics against targets that were previously difficult or deemed intractable to modulate.”
Despite the incredible progress the organizations have made, however, the process has involved numerous challenges. The rapid growth of artificial intelligence technology and foundation models has introduced a variety of reliability and trust concerns as well— especially as they are increasingly being used for crucial applications. Pushmeet Kohli, who founded the Reliability Team at DeepMind and is now Vice President of Research at the company, explains that although some of the largest problems in the natural sciences can be solved by machine learning and AI, there is also a lot of time and resources that are invested to ensure that the models are trained and developed with the utmost care. Specifically, Kohli explains that he is constantly thinking about how to make the system more reliable at producing consistent and accurate results, and more importantly, ways to enable safety measures that recognize when the systems are at fault.
Undoubtedly, although Google and DeepMind are certainly not the only innovators in this arena, they are likely among the most mature players in the market. Scientists and developers globally have recognized that artificial intelligence has immense potential for this field and are rapidly working to create their own products. For example, a startup founded at the University of Michigan, Genomenon, has developed AI products to leverage genomic data to support precision diagnostics and therapeutics. Facebook’s parent company Meta also invested significant resources in this sector and developed its own basic protein-folding model named ESMFold. As of March 2023, the platform states that its latest update can predict nearly 772 million protein structures.
The lack of numerous competitors to AlphaFold and significant progress by other market figures is indicative of just how challenging this work is, especially when a high bar for safety and reliability is established. Nevertheless, despite the challenges, this field has the potential to create substantial impact in the world of medicine and healthcare.
