Posted on by Lawrence Tabak, D.D.S., Ph.D.
Each year, more than 2.8 million people in the United States develop bacterial infections that don’t respond to treatment and sometimes turn life-threatening . Their infections are antibiotic-resistant, meaning the bacteria have changed in ways that allow them to withstand our current widely used arsenal of antibiotics. It’s a serious and growing health-care problem here and around the world. To fight back, doctors desperately need new antibiotics, including novel classes of drugs that bacteria haven’t seen and developed ways to resist.
Developing new antibiotics, however, involves much time, research, and expense. It’s also fraught with false leads. That’s why some researchers have turned to harnessing the predictive power of artificial intelligence (AI) in hopes of selecting the most promising leads faster and with greater precision.
It’s a potentially paradigm-shifting development in drug discovery, and a recent NIH-funded study, published in the journal Molecular Systems Biology, demonstrates AI’s potential to streamline the process of selecting future antibiotics . The results are also a bit sobering. They highlight the current limitations of one promising AI approach, showing that further refinement will still be needed to maximize its predictive capabilities.
These findings come from the lab of James Collins, Massachusetts Institute of Technology (MIT), Cambridge, and his recently launched Antibiotics-AI Project. His audacious goal is to develop seven new classes of antibiotics to treat seven of the world’s deadliest bacterial pathogens in just seven years. What makes this project so bold is that only two new classes of antibiotics have reached the market in the last 50 years!
In the latest study, Collins and his team looked to an AI program called AlphaFold2 . The name might ring a bell. AlphaFold’s AI-powered ability to predict protein structures was a finalist in Science Magazine’s 2020 Breakthrough of the Year. In fact, AlphaFold has been used already to predict the structures of more than 200 million proteins, or almost every known protein on the planet .
AlphaFold employs a deep learning approach that can predict most protein structures from their amino acid sequences about as well as more costly and time-consuming protein-mapping techniques.
In the deep learning models used to predict protein structure, computers are “trained” on existing data. As computers “learn” to understand complex relationships within the training material, they develop a model that can then be applied for making predictions of 3D protein structures from linear amino acid sequences without relying on new experiments in the lab.
Collins and his team hoped to combine AlphaFold with computer simulations commonly used in drug discovery as a way to predict interactions between essential bacterial proteins and antibacterial compounds. If it worked, researchers could then conduct virtual rapid screens of millions of new synthetic drug compounds targeting key bacterial proteins that existing antibiotics don’t. It would also enable the rapid development of antibiotics that work in novel ways, exactly what doctors need to treat antibiotic-resistant infections.
To test the strategy, Collins and his team focused first on the predicted structures of 296 essential proteins from the Escherichia coli bacterium as well as 218 antibacterial compounds. Their computer simulations then predicted how strongly any two molecules (essential protein and antibacterial) would bind together based on their shapes and physical properties.
It turned out that screening many antibacterial compounds against many potential targets in E. coli led to inaccurate predictions. For example, when comparing their computational predictions with actual interactions for 12 essential proteins measured in the lab, they found that their simulated model had about a 50:50 chance of being right. In other words, it couldn’t identify true interactions between drugs and proteins any better than random guessing.
They suspect one reason for their model’s poor performance is that the protein structures used to train the computer are fixed, not flexible and shifting physical configurations as happens in real life. To improve their success rate, they ran their predictions through additional machine-learning models that had been trained on data to help them “learn” how proteins and other molecules reconfigure themselves and interact. While this souped-up model got somewhat better results, the researchers report that they still aren’t good enough to identify promising new drugs and their protein targets.
What now? In future studies, the Collins lab will continue to incorporate and train the computers on even more biochemical and biophysical data to help with the predictive process. That’s why this study should be interpreted as an interim progress report on an area of science that will only get better with time.
But it’s also a sobering reminder that the quest to find new classes of antibiotics won’t be easy—even when aided by powerful AI approaches. We certainly aren’t there yet, but I’m confident that we will get there to give doctors new therapeutic weapons and turn back the rise in antibiotic-resistant infections.
 2019 Antibiotic resistance threats report. Centers for Disease Control and Prevention.
 Benchmarking AlphaFold-enabled molecular docking predictions for antibiotic discovery. Wong F, Krishnan A, Zheng EJ, Stark H, Manson AL, Earl AM, Jaakkola T, Collins JJ. Molecular Systems Biology. 2022 Sept 6. 18: e11081.
 Highly accurate protein structure prediction with AlphaFold. Jumper J, Evans R, Pritzel A, Kavukcuoglu K, Kohli P, Hassabis D., et al. Nature. 2021 Aug;596(7873):583-589.
 ‘The entire protein universe’: AI predicts shape of nearly every known protein. Callaway E. Nature. 2022 Aug;608(7921):15-16.
Antimicrobial (Drug) Resistance (National Institute of Allergy and Infectious Diseases/NIH)
Collins Lab (Massachusetts Institute of Technology, Cambridge)
AlphaFold (Deep Mind, London, United Kingdom)
NIH Support: National Institute of Allergy and Infectious Diseases; National Institute of General Medical Sciences
Posted on by Dr. Francis Collins
These round, multi-colored orbs in the illustration above may resemble SARS-CoV-2, the coronavirus responsible for COVID-19. But they’re actually lab-made nanocrystals called quantum dots. They have been specially engineered to look and, in some ways, act like the coronavirus while helping to solve a real challenge for many labs that would like to study SARS-CoV-2.
Quantum dots, which have been around since the mid-1980s, are designed with special optical properties that allow them to fluoresce when exposed to ultraviolet light. The two pictured here are about 10 nanometers in diameter, about 3,000 times smaller than the width of a human hair. The quantum dot consists of a semi-conductive cadmium selenide inner core (orange) surrounded by a zinc sulfide outer shell (teal). Molecules on its surface (yellow) allow researchers to attach the viral spike protein (purple), which SARS-CoV-2 depends on to infect human cells.
To the left is a human cell (gray) studded with the ACE2 receptors (blue) that those viral spike proteins bind to before SARS-CoV-2 enters and infects our cells. In the background, you see another spike protein-studded quantum dot. But human neutralizing antibodies (pink) are preventing that one from reaching the human cell.
Because SARS-CoV-2 is so highly infectious, basic researchers without access to specially designed biosafety facilities may be limited in their ability to study the virus. But these harmless quantum dots offer a safe workaround. While the quantum dots may bind and enter human cells just like the virus, they can’t cause an infection. They offer a quick, informative way to assess the potential of antibodies or other compounds to prevent the coronavirus from binding to our cells.
In work published in the journal ACS Nano, a team that included Kirill Gorshkov, NIH’s National Center for Advancing Translational Sciences (NCATS), Rockville, MD, along with Eunkeu Oh and Mason Wolak, Naval Research Laboratory, Washington, D.C., demonstrated how these quantum dots may serve as a useful new tool to speed the search for new COVID-19 treatments. The dots’ fluorescent glow enabled the researchers to use a microscope to observe how these viral mimics bind to ACE2 in real time, showing how SARS-CoV-2 might attach to and enter our cells, and suggesting ways to intervene.
Indeed, imagine thousands of tiny wells in which human cells are growing. Imagine adding a different candidate drug to each well; then imagine adding the loaded quantum dots to each well and using machine vision to identify the wells where the dots could not enter the cell. That’s not science fiction. That’s now.
With slightly different versions of their quantum dots, the NCATS researchers and their colleagues at the Naval Research Laboratory will now explore how other viral proteins are important for the coronavirus to infect our cells. They also can test how slight variations in the spike protein may influence SARS-CoV-2’s behavior. This work provides yet another stunning example of how scientists with widely varying expertise have banded together—using all the tools at their disposal—to forge ahead to find solutions to COVID-19.
 Quantum dot-conjugated SARS-CoV-2 spike pseudo-virions enable tracking of angiotensin converting enzyme 2 binding and endocytosis. Gorshkov K, Susumu K, Chen J, Xu M, Pradhan M, Zhu W, Hu X, Breger JC, Wolak M, Oh E. ACS Nano. 2020 Sep 22;14(9):12234-12247.
What are Quantum Dots? (National Institute of Biomedical Imaging and Bioengineering/NIH)
Coronavirus (COVID-19) (NIH)
I Am Translational Science: Kirill Gorshkov (National Center for Advancing Translational Sciences/NIH)
U. S. Naval Research Laboratory (Washington, D.C.)
NIH Support: National Center for Advancing Translational Sciences
Posted on by Dr. Francis Collins
In 1953, Francis Crick famously told the surprised customers at the Eagle and Child pub in London that he and Jim Watson had discovered the secret of life. When NIH’s Marshall Nirenberg and his colleagues cracked the genetic code in 1961, it was called the solution to life’s greatest secret. Similarly, when the complete human genome sequence was revealed for the first time in 2003, commentators (including me) referred to this as the moment where the book of life for humans was revealed. But there are many more secrets of life that still need to be unlocked, including figuring out the biochemical rules of a protein shape-shifting phenomenon called allostery .
Among those taking on this ambitious challenge is a recipient of a 2018 NIH Director’s New Innovator Award, Srivatsan Raman of the University of Wisconsin-Madison. If successful, such efforts could revolutionize biology by helping us better understand how allosteric proteins reconfigure themselves in the right shapes at the right times to regulate cell signaling, metabolism, and many other important biological processes.
What exactly is an allosteric protein? Proteins have active, or orthosteric, sites that turn the proteins off or on when specific molecules bind to them. Some proteins also have less obvious regulatory, or allosteric, sites that indirectly affect the proteins’ activity when outside molecules bind to them. In many instances, allosteric binding triggers a change in the shape of the protein.
Allosteric proteins include oxygen-carrying hemoglobin and a variety of enzymes crucial to human health and development. In his work, Raman will start by studying a relatively simple bacterial protein, consisting of less than 200 amino acids, to understand the basics of how allostery works over time and space.
Raman, who is a synthetic biologist, got the idea for this project a few years ago while tinkering in the lab to modify an allosteric protein to bind new molecules. As part of the process, he and his team used a new technology called deep mutational scanning to study the functional consequences of removing individual amino acids from the protein .
The screen took them on a wild ride of unexpected functional changes, and a new research opportunity called out to him. He could combine this scanning technology with artificial intelligence and other cutting-edge imaging and computational tools to probe allosteric proteins more systematically in hopes of deciphering the basic molecular rules of allostery.
With the New Innovator Award, Raman’s group will first create a vast number of protein mutants to learn how best to determine the allosteric signaling pathway(s) within a protein. They want to dissect out the properties of each amino acid and determine which connect into a binding site and precisely how those linkages are formed. The researchers also want to know how the amino acids tend to configure into an inactive state and how that structure changes into an active state.
Based on these initial studies, the researchers will take the next step and use their dataset to predict where allosteric pathways are found in individual proteins. They will also try to figure out if allosteric signals are sent in one direction only or whether they can be bidirectional.
The experiments will be challenging, but Raman is confident that they will serve to build a more unified view of how allostery works. In fact, he hopes the data generated—and there will be a massive amount—will reveal novel sites to control or exploit allosteric signaling. Such information will not only expand fundamental biological understanding, but will accelerate efforts to discover new therapies for diseases, such as cancer, in which disruption of allosteric proteins plays a crucial role.
 Allostery: an illustrated definition for the ‘second secret of life.’ Fenton AW. Trends Biochem Sci. 2008 Sep;33(9):420-425.
 Engineering an allosteric transcription factor to respond to new ligands. Taylor ND, Garruss AS, Moretti R, Chan S, Arbing MA, Cascio D, Rogers JK, Isaacs FJ, Kosuri S, Baker D, Fields S, Church GM, Raman S. Nat Methods. 2016 Feb;13(2):177-183.
Drug hunters explore allostery’s advantages. Jarvis LM, Chemical & Engineering News. 2019 March 10
Allostery: An Overview of Its History, Concepts, Methods, and Applications. Liu J, Nussinov R. PLoS Comput Biol. 2016 Jun 2;12(6):e1004966.
Srivatsan Raman (University of Wisconsin-Madison)
Raman Project Information (NIH RePORTER)
NIH Director’s New Innovator Award (Common Fund/NIH)
NIH Support: National Institute of General Medical Sciences; Common Fund
Posted on by Dr. Francis Collins
When you think of the causes of infectious diseases, what first comes to mind are probably viruses and bacteria. But parasites are another important source of devastating infection, especially in the developing world. Now, NIH researchers and their collaborators have discovered a new kind of treatment that holds promise for fighting parasitic roundworms. A bonus of this result is that this same treatment might work also for certain deadly kinds of bacteria.
The researchers identified the potential new therapeutic after testing more than a trillion small protein fragments, called cyclic peptides, to find one that could disable a vital enzyme in the disease-causing organisms, but leave similar enzymes in humans unscathed. Not only does this discovery raise hope for better treatments for many parasitic and bacterial diseases, it highlights the value of screening peptides in the search for ways to treat conditions that do not respond well—or have stopped responding—to more traditional chemical drug compounds.
Posted on by Dr. Francis Collins
When Dmitry Lyumkis headed off to graduate school at The Scripps Research Institute, La Jolla, CA, he had thoughts of becoming a synthetic chemist. But he soon found his calling in a nearby lab that imaged proteins using a technique known as single-particle cryo-electron microscopy (EM). Lyumkis was amazed that the team could take a purified protein, flash-freeze it in liquid nitrogen, and then fire electrons at the protein, capturing the resulting image with a special camera. Also amazing was the sophisticated computer software that analyzed the raw 2D camera images, merging the data and reconstructing it into 3D representations of the protein.
The work was profoundly complex, but Lyumkis thrives on solving extremely difficult puzzles. He joined the Scripps lab to become a structural biologist and a few years later used single-particle cryo-EM to help determine the atomic structure of a key protein on the surface of the human immunodeficiency virus (HIV), the cause of AIDS. The protein had been considered one of the greatest challenges in structural biology and a critical target in developing an AIDS vaccine .
Now, Lyumkis has plans to take single-particle cryo-EM to a whole new level—literally. He wants to develop new methods that allow it to model the atomic structures of much smaller proteins. Right now, single-particle cryo-EM has worked with proteins as small as roughly 150 kilodaltons, a measure of a protein’s molecular weight (the approximate average mass of a protein is 53 kDa). Lyumkis plans to drop that number well below 100 kDa, noting that if his new methods work as he hopes, there should be very little, if any, lower size limit to get the technique to work. He envisions generating within a matter of days or weeks the precise structure of an average-sized protein involved in a disease, and then potentially handing it off as an atomic model for drug developers to target for more effective treatment.