‘Google Pharma’: A New Approach to Drug Discovery?

Author: Aditi Somayajula, Graphics: Acasia Giannakouros

The BRB Bottomline

In a major scientific breakthrough, Google-owned research lab DeepMind solved the decades-long protein folding problem with AI. The science is cutting-edge and suggests that big tech might become the next big pharma.

It is no exaggeration to claim that proteins are the building blocks of life. Found at the forefront of nearly every cellular task, proteins conduct our immune response, repair damage to tissues, equip us with the five senses, and regulate countless processes in the body. Not surprisingly, many of the world’s scientific challenges, from understanding disease to developing drugs, are fundamentally protein-related problems. Each stride towards discovering the molecular workings of proteins marks a significant step in our ability to understand and redesign biology. 

A Blueprint for Proteins

A Biological Context

Every protein begins as a long one-dimensional chain of amino acids that are joined together in a distinct order. In order to become functional, amino acid chains transform into complex three-dimensional structures via many stages of folding. This three-dimensional architecture is critical as it endows the protein with a highly specific function—for instance, antibodies fold into unique shapes that allow them to identify and destroy invading pathogens in key-lock fashion. Properly folded proteins yield effective cellular processes whereas many of the most tragic diseases can be attributed to proteins that do not form or function as expected.

The Protein Folding Problem

In 1972, Nobel laureate Christian Anfinsen postulated that the amino acid sequence of a protein—its one-dimensional blueprint—should fully explain its shape and structure. For decades, unlocking the secrets of the form and function of proteins was limited to extensive trial and error in laboratories and was the product of hours of tedious work and multimillion dollar equipment. But proving Anfinsen’s prediction right remained pertinent throughout the latter part of the 20th century and well into the current technological and digital era. Very recently, new developments in technology have allowed scientists to hypothesize a computational approach to predicting the architecture of a protein in place of expensive laboratory alternatives. 

But what made devising a computational approach particularly difficult was a baffling aspect of the protein folding problem known as Levinthal’s paradox. In nature, proteins achieve their final folding arrangement within a few milliseconds. This is quite astounding given that the number of ways a protein could theoretically fold before achieving its final configuration is tremendous — an estimated 10300 possibilities per protein. Manually computing all folding arrangements via a brute force approach would take longer than the age of the known universe

However, despite being nightmarishly difficult, understanding the folding mechanisms of a protein is crucial to understanding our own biology—and it is of little surprise that solving the protein folding problem has been somewhat of a holy grail in the scientific community. 

A Scientific Breakthrough

Late last year, it appeared that an elegant solution to the protein folding problem was realized. Google-owned artificial intelligence research lab DeepMind announced a major breakthrough in the protein folding problem via their solution known as AlphaFold

Why Neural Networks Work

Based in London, DeepMind is a leader in the development of neural networks, a form of artificial intelligence that is modeled after networks of neurons in the human brain. Akin to how the human brain learns to recognize patterns—such as of familiar faces or locations—neural networks ‘learn’ to recognize patterns in large datasets and eventually become ‘intelligent’ enough to draw inferences. In a biological context, AlphaFold’s neural networks were presented with large datasets of one-dimensional amino acid sequences, their consequent three-dimensional protein structures, and their physical and geometric space constraints. Over time, the AI detected patterns in how certain amino acid chains tended to fold into distinct proteins. AlphaFold’s neural networks were then able to utilize such patterns to make predictions about how new amino acid sequences would fold into proteins, an endeavor that had long eluded laboratory attempts. 

When AI Triumphs

In late 2020, Google’s AlphaFold entered a protein structure prediction competition known as CASP (Critical Assessment of Structure Prediction) in which participants submit folding predictions of particular amino acid sequences. Folding predictions are then compared against known structures that are experimentally determined. AlphaFold’s stunning performance at CASP was considered to be an extraordinary breakthrough, significantly outperforming its competition and eclipsing any other attempts to solve the protein folding problem. Although the AI was not perfect, its margin of error was quite remarkable. AlphaFold successfully predicted the three-dimensional architecture of proteins to the width of a single atom. Nearly one year later, AlphaFold’s AI has allowed us to obtain the protein structures of 98.5% of the human proteome; before AlphaFold, our documentation hovered somewhere around 17%. 

Reinventing Research

AlphaFold’s breakthrough technology allows us to better understand how the delicate folding structure of a protein interacts with cells to spur life and disease alike. Last July, the potential for a better understanding of numerous diseases exploded when DeepMind decided to open-source AlphaFold so that its program code could be accessed online by anyone, anywhere. Now, biologists around the world are able to freely contribute to protein-related disease research and drug development. Not surprisingly, an entire ecosystem of biology research remains to be discovered. 

Early uses of AlphaFold seem highly promising. Recently, researchers at the University of California, San Francisco utilized open-source AlphaFold to aid in the development of therapeutics for COVID-19. Similarly, a research group at the University of Colorado, Boulder used AlphaFold to understand why certain bacterial strains develop a resistance to antibiotics. A challenge that had once eluded the research team for years, AlphaFold enabled them to solve the problem in a mere fifteen minutes. 

A New Approach to Drug Discovery

Drug development involves identifying a particular target protein in the body such that a particular drug can latch to the protein with the intent of producing a favorable health outcome. Prior to AlphaFold, the research process behind placing a drug on the market was exceptionally tedious—20,000 human genes can malfunction in roughly 100,000 ways and spur millions of unique interactions between proteins. Unsurprisingly, roughly only one in every ten drugs in clinical trials made it onto a pharmacy shelf. 

Broadly, AlphaFold gives scientists the opportunity to understand the three-dimensional origami of a target protein without need for costly laboratory equipment, or maybe decades worth of research. Currently, research groups across the world are using AlphaFold to better understand the proteins behind particular rare diseases, certain forms of cancer, dementia, COVID-19, Alzheimer’s disease, Parkinson’s disease, and cystic fibrosis

Google Pharma?

From a business standpoint, Google’s acquisition of DeepMind speaks to a general trend of technology giants entering the healthcare market. Just this month, Alphabet, the parent company of Google, launched Isomorphic Labs, a spinoff that aims to reinvent the drug discovery process with an AI approach. As the pharmaceutical industry is incredibly lucrative at roughly a $1.27 trillion valuation, even getting a sliver of the market would be a tremendous success. Additionally, mega technology companies like Google already have a large network of shareholders and the resources to make an aggressive entrance into new markets. 

From a digital standpoint, technology giants likely have something to offer as many aspects of the drug discovery process have yet to be streamlined by tech savvy. Their entrance into the healthcare market will likely challenge existing big pharma to innovate. But the implications of technology companies joining the medical space might be something to consider before overstating their potential. For one, the creation of ‘Google Pharma’ or ‘Amazon Pharma’ will make yet another aspect of users’ daily lives dependent on these companies. Additionally, the existing business model could not transcend — while many of Google’s services, such as its search engine, are freely available to everyone, in many ways, we pay with our privacy. And in the healthcare space, the cost of a privacy breach is considerably steeper

As for Isomorphic Labs specifically, their aim to reimagine drug discovery might not translate to drugs hitting the market as quickly as some may hope, even if they leverage the use of AlphaFold. While AlphaFold undoubtedly democratizes and facilitates drug discovery, drug research and development is only the first step in the Byzantine process of testing, clinical trials, manufacturing, marketing, and distribution. In other words, until Google also decides to start manufacturing pills and needles themselves, all but the first step of creating a successful new drug would remain at the mercy of existing big pharma companies. 

A Final Word

It is inarguable that Google’s DeepMind should be credited for a major scientific breakthrough. Due to cutting-edge innovation in AI and proteomics, researchers are that much closer to discovering cures to diseases whose molecular explanations once eluded us. But before turning to tech giants as the new big pharma, it is important to consider the implications of such companies entering the medical space. 

Transforming the fundamental insights of AlphaFold into drugs that improve the lives of people will certainly require years of hard work from researchers, medical professionals, and technology professionals alike. But while the full impact of AlphaFold and its ability to transform drug discovery is yet to be assessed, there is no doubt that we are at the forefront of an exciting medical revolution with the enormous potential to change lives around the world. 

Take-Home Points

  • Last year, Google-owned DeepMind solved the decades-long protein folding problem with an elegant AI solution known as AlphaFold.
  • AlphaFold has fundamentally improved our understanding of the human proteome and is informing new methods of drug discovery. 
  • While AI will inevitably revolutionize medicine, it is important to consider the implications of big tech becoming the next pharma. 
  • Although its total impact is not fully understood, AlphaFold has undoubtedly led us that much closer to curing countless diseases. 

Leave a Reply

Your email address will not be published.