Future Medicine: Physics, Biology, And AI Will Transform Human Health

Co-written by Chuck Brooks and Dr. Thomas A. Cellucci, MBA

Verticals that will be most impacted by innovative developments in technology and science are the disciplines of medicine, biotechnology, and health. Those industry verticals will see a profound growth of technological innovation in the near future.

Twenty years ago, Craig Venter and Daniel Cohen remarked, “If the 20^th century was the century of physics, the 21st century will be the century of biology.” Since then, there have been some amazing advances in the fields of biotechnology and bioscience, with the promise of even more astounding breakthroughs to come. Over the past decade, we have seen significant strides in artificial intelligence, with radical long-term implications for every human endeavor. And now the convergence of the fields of physics, biology, and AI promises a far greater impact on humanity than any one of these fields alone. Even though a path to successfully integrating these fields exists, it is neither easy nor clear cut—but if done correctly, will revolutionize medicine and human health.

The Human Genome Project was just the beginning of the journey to find treatments for human diseases. Our genes are merely a codebook for making different proteins. These proteins are the fundamental building blocks for our cell structures and are responsible for their core functions. Understanding the biological processes behind a disease means identifying the specific protein or proteins whose undesirable effects in the body cause that disease. To treat the disease, a therapeutic agent is needed. This is usually given as a pill or an injection and contains active drug molecules that bind to copies of a disease-associated protein and change how they work.

4 key steps associated with finding treatments for human disease are as follows:

1) Figure out which proteins are implicated in the disease

2) Ensure that we can produce those proteins in a form conducive to testing and determine their structures either experimentally or through computer models

3) Find binders to those proteins (these binders are the drug candidates, and this step is the most important)

4) Test those binders first in vitro, then advance the most promising ones to animal tests, and finally select the best ones for clinical trials in humans

Step 1: Identifying Target Proteins

There has been tremendous amount of work in labs worldwide over the last three decades to identify the proteins implicated in various diseases. This work will continue into the near future, and AI-driven knowledge graphs can complement and speed up the work of biologists.

To map and characterize all the proteins encoded and produced by the human genome, scientists have embarked on the Human Proteome Project (HPP). So far, HPP has characterized 18,397 out of the estimated 19,778 proteins that make up the human body.[1] However, it is estimated that only 10,248 play roles in human disease.[2] This subset of proteins is what scientists call the “druggable” proteome.

Step 2: Isolating Target Proteins & Determining Their Structure

To verify whether a drug candidate works, it must be tested against a target protein implicated in a disease. Isolating target proteins is important to ensure target proteins are available for testing against drug candidates and for determining their three-dimensional structures. Some proteins are easy to isolate and purify; others can only be produced in conjunction with their associated cellular machinery. Scientists have made tremendous strides in determining the three-dimensional structures of these target proteins through x-ray crystallography and NMR. These structures are freely available for all to use in databases such as the Protein Data Bank (PDB). Now, AI is helping figure out the structures of those proteins for which we have not yet experimentally validated structures.

Step 3: Finding Drug Candidates

The Most Crucial Step

Once the target protein is properly characterized, then begins the truly daunting challenge of searching for drug-like molecules that bind to the protein and effectively intervene in the disease process. All of the work in steps 1 and 2 is for naught if we cannot find drug candidates that bind to the target proteins. So-called small-molecule drugs are the mainstay of modern medicine. Most pharmaceuticals are small molecules. Not only can these low-molecular-weight organic molecules bind to disease-associated proteins outside of cells, but they can also pass through cell membranes to bind to target proteins inside cells. Typically administered orally as shelf-stable drugs, small-molecule pharmaceuticals can be also manufactured and distributed efficiently.

A Challenging Process

The process of finding small-molecule drug candidates primarily relies on trial and error. The isolated and purified protein is tested against a collection of previously synthesized compounds stored in the lab to check for signs of a chemical reaction in the hope of finding a binder. But there is a fundamentally crippling limitation to this approach. Over the past 150 years, humanity has made fewer than 10 million distinct drug-like small-molecule structures, or chemotypes. Yet estimates of the total number of unique drug-like small-molecule chemotypes possible under the rules of chemistry range from a decillion (that is, 10³³, or a one followed by 33 zeroes) to a novemdecillion (10⁶⁰, or a one followed by 60 zeroes).

Even if it were possible to magically accelerate the creation of brand-new chemical structures to one per second, it would take more than three septillion (that is, three plus 25 zeros) years to create a decillion chemotypes. That is well over two trillion times the currently estimated age of the universe! In other words, figuring out which small molecules would make viable drugs through brute-force synthesis and testing is an utterly impossible task.

Some companies have turned to AI as a substitute for trial-and-error experimentation. But can AI really help in the field of small-molecule drug discovery? AI only produces solutions similar to the ones for which it has seen examples in the data used to train it. Expecting AI to find drug candidates for which there is no pre-existing data is even more unreasonable than expecting Chat GPT trained exclusively on massive amounts of English webpage text to suddenly construct sentences in perfect Russian. AI will find drug-like molecules similar to the fewer than 10 million chemotypes for which experimental data already exists, but it cannot explore the other decillion+ possibilities. AI trained on existing data cannot unlock the secrets of a decillion unexamined compounds.

What if pharmaceutical companies could design never-before-made molecules on the computer and figure out if they will bind to a target protein without having to actually make the molecule, isolate the protein, and run experiments? We already rely on such technologies every day in other areas. For example, no one erects reduced-scale models of buildings or bridges to see if they will stand before actually constructing them.

But modeling the quantum physics of how a drug molecule binds to a protein in water is a forbiddingly complex challenge. Attempting to model the interaction between a protein with thousands of atoms and a drug-like molecule with hundreds of atoms quickly becomes intractable, exceeding the computing power of even the most advanced computing clouds. To tackle this extreme level of complexity, most simulation-based drug discovery technology companies found that drastic compromises were necessary. These compromises involved crude approximations of physics and computational short cuts. And invariably, these compromises led to gross inaccuracies when modeling molecular interactions.

Are both AI-based and physics-based approaches dead ends? Not quite. Various companies are still trying both approaches. However, reaching sufficient accuracy requires significant fundamental advances in modeling the physics of protein-drug interactions. Pfizer’s former Senior VP of R&D Strategy, Robert Karr, points out that one company, Verseon, has built a seamless platform incorporating propriety advances in physics and AI. In his words, “Verseon’s disruptive platform changes how drugs can be discovered and developed, and the company is poised to make a dramatic impact on modern medicine.” After examining the company’s drug-discovery platform, Nobel Laureate Hartmut Michel said, “The fundamental advancements Verseon has made in quantum mechanical modeling of protein-drug interactions are extremely impressive.”

Verseon has built technology to design never-before-made novel drug molecules on the computer. The company’s advances in quantum-physics modeling, which it calls Deep Quantum Modeling™ (DQM™), are sufficiently accurate to determine whether its computationally designed new drug-like molecules will bind to a target protein before making and testing them in the lab. And AI trained on data from these new molecules identified by DQM will help generate more variants from which to pick the best ones for clinical trials.

Step 4: In Vitro, Preclinical, and Clinical Testing

All new drug candidates must still be taken through preclinical and clinical trials to fully establish their safety and effectiveness. This process typically takes years to complete, and process improvements that speed promising new treatments to market would be a welcome boon. AI can help with regulatory documentation. Weave is one example of a company selling solutions that utilize AI to streamline the process of preparing Investigational New Drug regulatory filings (INDs). Preparing INDs is one area that could benefit from automation. The company touts its ability to create INDs faster without sacrificing quality and offers greater control over data organization for IND filings.

Human organoid-based testing during the preclinical process is an emerging technique that can dramatically improve the characterization of how a drug will behave in clinical trials. Organoids are miniature versions of organs in the human body that are generated using adult stem cells. After taking samples from patients and trial volunteers, these lab-grown replicas mimic the function of various organs. Testing drug candidates on organoids can give a lot of clues into how the human body would react to a drug candidate before human trials.

Regulators like the US FDA are tasked with weeding out unsafe and ineffective drug candidates that enter trial pipelines. But it is a process that takes years once drug candidates enter the clinic. The long wait for approval frustrates both companies eager to enter the market with innovative products and patients desperate for a solution to a serious medical condition. The need to accelerate the process is not lost on regulators. The US FDA has set up processes such as Fast Track and Breakthrough Therapy designations to accelerate the clinical trial process for novel drugs that show uniquely desirable therapeutic properties.

The Implications of Using Advances in Physics, Biology, and AI to Their Full Potential

Assuming the pharmaceutical development ecosystem takes full advantage and coordinates the use of advances like those described above, what could medicine achieve?

Adityo Prakash and Tom Cellucci point out some of the implications in their recent article appearing in Homeland Security Today. They approach the subject from the perspective of both industry and national security. Adityo Prakash is the CEO of Verseon. He confounded Verseon to change how the world finds new medicines. Tom Cellucci drove America’s nanotechnology agenda under presidents Clinton and George W. Bush, and he served as the US Federal Government’s first-ever Chief Commercialization Officer at the US Department of Homeland Security (DHS) under both the Bush and Obama administrations. Prakash and Cellucci point out that current medicine has so far only uncovered drug candidates that bind 670 of the druggable proteins in the human body. They contend that better methods for finding drug candidates and streamlining required testing will accelerate the march toward being able to address all 10,248 druggable proteins. Once this goal is reached, medicine will be able to intervene with exquisite precision in disease processes that inexorably progress over the human lifespan. Being able to precisely control the entire druggable set of proteins in the human body will also facilitate agile therapeutic development as a first line of defense against new infectious agents that try to enter our cells.

Emerging medical and life science technologies are helping contribute to an inward evolution. Scientists are learning from the exponentially growing data on our health and disease states using data analytics and AI-driven insights generated from such data. They are developing a deeper understanding of the concert of proteins involved in the progression of disease and the best pathways for both treatment and prevention.

Given that most major diseases are associated with aging, regulating the functions of proteins in our bodies will allow medicine to preserve a youthful and disease-free state far longer than is possible today—and eventually even to partially rejuvenate our bodies from the cellular level. These advances may make it possible to turn 80, 90, or even 100 into the new 50. But even before such dramatic increases in human health span happen, progress toward addressing large subsets of the druggable proteome will deliver a steady flow of amazing new drugs. These drugs will treat a great many diseases far better than we can today. Then the 21st century will indeed establish itself as the century of biology.