Discovery is the most crucial stage of drug development because it has the potential to save pharmaceutical companies billions of dollars in wasted time and money on drug candidates that are unlikely to succeed. Artificial intelligence (AI) is an essential capability becoming more prevalent in solving challenges at this stage, enabling scientists to identify potential drug targets faster, screen more compounds in less time for activity against these targets, and find leads that have the best chance of progressing to become approved treatments.
While mainly focused on speeding up drug discovery, the latest research involving AI in drug discovery helps us better understand how diseases develop and progress, as well as drugs’ effects on humans. The more AI is used, the more it improves, so that better predictions are made in identifying potential new drugs.
Staying on target
The target identification step of drug discovery is where researchers pinpoint exactly what leads a certain condition or disease to progress. Computer modeling provides a big step forward in finding appropriate targets for drugs, and making that identification faster.
One example of recent research involving such computer modeling comes from the University of California San Diego School of Medicine. Researchers modeled lipoprotein-associated phospholipase A2 (Lp-PLA2), an enzyme important to cardiovascular health, to learn more about how Lp-PLA2 interacts with the phospholipid membrane, something that’s not currently well-known, according to a press release on the research. Understanding this interaction could lead to better knowledge of how cardiovascular disease develops and lead to new, more effective treatments for the disease. The research was published in early January 2022 in PNAS.
“I am very pleased that we were able to go into much greater depth on how this enzyme works than ever before,” said Edward A. Dennis, PhD, senior author of the study, in the press release. “Using the latest advances in lipidomics and computational molecular dynamics simulations, we got a picture which is worth a thousand words. We now have movies that show how this enzyme works at the atomic level, and that should help us figure out ways to activate or inactivate the enzyme as necessary for health.”
In other target ID-related work, researchers from Queen’s University Belfast recently developed a computer modeling tool that better predicts new, more selective binding sites for possible drugs, leading to better drug targeting and more effective drugs. In particular, their tool involves a novel class of compounds known as allosteric drugs in G protein-coupled receptors (GPCRs).
According to a press release on the research, GPCRs are the biggest membrane protein group that converts signals inside cells from a variety of endogenous molecules such as neurotransmitters and hormones. Due to their wide effect on human physiology, they are a key target of many drugs. However, discovering GPCR drugs is difficult because they tend to bind to more than one protein target, leading to unintended side effects. Other research highlights allosteric sites as alternative binding sites for drugs, but they are difficult to identify. In the Queen’s University Belfast research, published in September 2021 in ACS Central Science, scientists used certain probes from G protein-coupled receptor allosteric ligands to more easily and accurately identify these allosteric sites than current methods.
“Our pipeline can identify allosteric sites in a short time, which makes it suitable for industry settings,” said senior author Dr. Irina Tikhonova in a press release about the study. “As such, our pipeline is a feasible solution to initiate structure-based search of allosteric drugs for any membrane-bound drug targets that have an impact on cancer, inflammation, and CNS [central nervous system] diseases.”
Screening at high speed
In recent work related to the compound screening stage of drug discovery, where scientists look at thousands of possible drug candidates for activity against identified targets, AI is fostering important advancements. This new technology continues to push the envelope of how many compounds can be examined in parallel, and how quickly the screening can be completed.
Research from scientists at the University of Southern California Dornsife College of Letters, Arts and Sciences that was published in Nature in December 2021 is just one recent example. They developed a new virtual screening method called V-SYNTHES (virtual synthon hierarchical enumeration screening) for screening that is significantly faster and less costly than current methods. Focused on readily available for synthesis (REAL) combinatorial libraries, the scientists used V-SYNTHES to screen 11 billion compounds for the best chance of having a good first fit with target proteins.
Such virtual libraries of compounds are growing at an exponential rate. Actually synthesizing all of the compounds to test fit in real life is impractical, a press release on the project points out, so computational tools are needed to narrow the list down to only compounds most likely to work. However, even with current computational tools, the process of going through these virtual libraries is still expensive and tedious. V-SYNTHES offers a solution to this problem.
Focusing specifically on novel cannabinoid antagonists, the compounds predicted by V-SYNTHES showed a 33 percent hit rate once they were synthesized and tested in real life, a substantial improvement over standard virtual screening of the same library, which needed about 100 times more computational resources, the authors write.
“V-SYNTHES first identifies the best scaffold–synthon [chemical building block] combinations as seeds suitable for further growth, and then iteratively elaborates these seeds to select complete molecules with the best docking scores,” the authors explain in the Nature study, adding that their approach can be scaled easily as virtual compound libraries continue to grow. According to the press release about their research, the scientists are now looking into further automating their method, as it currently needs plenty of human assistance.
While not directly involving AI, another interesting project published in Nature Photonics on Jan. 13 could also help speed the drug screening process. In their work, researchers at the University of Michigan and the University of Bath discovered a new optical effect caused by twisted nanoscale semiconductors—third-harmonic Mie scattering optical activity. According to a press release on the work, this effect could be used to further speed up high-throughput screening, which involves microplates with tiny wells containing a tiny sample of each compound so thousands can be analyzed at once. As these microwells get smaller and smaller to screen ever-increasing numbers of compounds at once, the optical effect could offer a new method to analyze the tiny amounts of samples they contain.
Preventing issues down the road
Other research related to AI aims to improve current models. In one recent example published in the Royal Society of Chemistry journal CrystEngComm, scientists merged published and proprietary datasets to provide better training for machine learning models, so those models could better predict a substance’s ability to crystalize for use in potential new drugs. Crystal structure is important in drug discovery. It “can help to rationalize conformational effects, for example, or characterize the chemistry of a new chemical entity where other techniques have led to ambiguity,” said study author Dr. Jason Cole in a press release. “Later in the process, when a new chemical entity is studied as a candidate molecule, crystal structures are critical as they inform form selection and can later aid in overcoming formulation and tableting issues.”
Developments going forward
While these are only some examples of recent advances relating to AI and computer science in drug discovery research, trends indicate that AI will continue to contribute to early-stage drug development going forward. A 2020 World Economic Forum article highlighting the many benefits of AI to drug discovery says that partnerships between biopharma and digital technology companies will likely become more common over this decade as companies aim to make drug development faster and more efficient.
However, the benefits of AI also come with challenges, particularly around the sheer amount of data as well as how quickly it’s growing and how diverse it is, which may leave current machine learning methods unable to cope, according to a 2021 review in Drug Discovery Today. Such challenges will likely only drive researchers to come up with ever more innovative AI methods to mine valuable insights from this expanding mountain of data, enabling further breakthroughs that could eventually become more effective treatments for patients around the world.