Soon after a team of researchers from Schrödinger published a commentary in Cell earlier this year detailing their model for predicting binding to human ether-a-go-go related gene (hERG), they sent a copy to Bill Gates, whose Bill and Melinda Gates (BMG) Foundation Trust is the company’s second largest shareholder.
“He was as excited as we thought he would be, as anybody would be, and we actually have been getting very, very positive responses to that paper,” Schrödinger President and CEO Ramy Farid, PhD, recalled, speaking with GEN Edge.
Among those responding positively was Trevor Mundel, PhD, who leads the Gates Foundation’s efforts to develop high-impact initiatives against leading causes of death and disability in developing countries as President, Global Health. Mundel had long been interested in the longtime biopharma challenge of predicting toxicity risk early in drug discovery, having contacted the FDA when it published its Predictive Toxicology Roadmap in 2017.
“That led to the conversation around, how do we work together on helping to fund the project, and helping with discussions with the FDA, and discussions with potential partners,” Farid said. “They’ve got a number of grantees that are trying to develop safe medicines, avoid having to do unnecessary animal studies. So, the collaboration was formed there.”
Those discussions led to Schrödinger recently receiving a $10 million grant from the Gates Foundation toward expanding its computational platform to predict toxicity risk early in drug discovery.
Schrödinger says it plans to use the funding toward developing a computational solution designed to reduce the risk of development failure associated with binding to off-target proteins, and thus improve the properties of drug development candidates, by developing computational methods for predicting the binding of molecules to a panel, to an array of off-targets—an approach known as predictive toxicology.
De-Risking programs
“The idea behind predictive toxicology is to de-risk programs before you start putting molecules into humans, certainly before you put them in humans, but also before you put them into animals, to predict the toxicity profile or the safety profile of a molecule as it pertains to binding to off targets,” Farid said.
Schrödinger has already applied its new structure-based methods for predicting binding to hERG, a gene known as KCNH2 that encodes for the protein Kv 11.1, a subunit of a potassium ion channel (IKr), which is important for cardiac repolarization. Dysfunction of hERG causes long QT syndrome and sudden death, which occur in patients with cardiac ischemia.
In addition to ion channels like hERG, another important class of off-targets to which Schrödinger is working to prevent binding is Cytochrome P450s. The company has also built structure-based models for Cytochrome P450 (CYP) 3A4, 2D6, and 2C9—a trio of off targets that are enzymes in the human liver that metabolize drugs and other substances, but which have been linked to adverse drug interactions, increased disease risk, and negative personality traits such as anxiety and impulsivity.
Another off target Schrödinger has worked to model is the pregnane X receptor (PXR), a key regulator of xenobiotic metabolism and disposition in the liver that can negatively impact drug efficacy and safety when activated by diverse compounds to elevate metabolism.
“When you’re just working on a project, you’ve got potentially 20,000 proteins in the human proteome that you can bind to. And in many cases, it’s not good to bind to any of those. It’s going to cause some kind of issue that you might not even fully understand,” Farid said.
Farid said Schrödinger’s predictive tox solution will encompass more than software, including potentially an artificial intelligence (AI) platform and the algorithms that would power it: “This is existing technology, already validated, already deployed on a number of these targets. Our project is industrializing this effort. It’s saying, ‘Look, we’ve gotten it to work on a handful of off targets. Let’s do many more of them.”
Aiming for 100
For how many off targets does Schrödinger envision developing models?
“We would really like in the next few years to be able to enable and develop a computational toxicology panel that includes approximately 100 off-targets,” Farid said. “That would be the goal, to develop all those models in, say, two to three years.”
Schrödinger’s predictive tox integrates both physics-based methods and machine learning (ML) to amplify those physics-based methods, an integration the company says is enabled by:
- Its computational platform reaching a level of accuracy enabling it to model off targets.
- The structural biology “revolution” of advances in experimental and computational methods for predicting the structures of proteins.
- The accelerating speed of computers. Schrödinger is using Nvidia’s AI technologies.
“This is not an ML-only solution. We’ve already tried that. That does not work,” Farid said.
Does Schrödinger envision its predictive toxicology initiative replacing traditional tox, or the pure machine learning approach employed by a growing number of companies?
“We believe it will replace, eventually, the ML-only methods, but it won’t replace ML,” Farid said. “You still need machine learning to be able to scale it to very large numbers of molecules, while you still need physics to be able to produce the training set. It will also not replace experimental methods completely. You obviously still have to test molecules. The idea is that you’ll be testing way fewer molecules experimentally in vitro.”
Karen Akinsanya, PhD, Schrödinger’s President of R&D, Therapeutics, added that the predictive tox initiative won’t eliminate the good laboratory practice (GLP) toxicology studies in animals that are now required and regulated by the FDA.
“Today, molecules tend to get pretty far along before we know much about their off-target binding to large numbers of proteins,” Akinsanya said. “If you know that really early on in a project, you’re going to save a lot of time and money. You won’t have to get to the very end and do all of these complex studies, and then have to U-turn and go back and start all over again.”
“Better molecules through discovery”
“This initiative really has the potential to get better molecules through discovery,” Akinsanya added.
Farid said Schrödinger plans to make its predictive tox solution available to its customers, “which is essentially the whole biopharmaceutical industry.”
Based in New York, Schrödinger has combined forms of AI with physics-based first principles for about two decades, in order to identify new drugs for targets that are designed to treat a variety of diseases—a key area where the company stands out, Farid and Akinsanya explained to GEN Edge last year.
Schrödinger’s business consists of two prongs:
- Licensing its software used in drug discovery and materials design, a business that has attracted some 1,750 customers to the company.
- Deploying its platform to drug discovery. Schrödinger maintains 13 active collaboration projects with biopharma and other partners focused on drug and materials design, with nine of those partners having advanced programs into the clinic.
Schrödinger has discussed its predictive toxicity effort with several of its biopharma customers, Farid said, though his company won’t disclose their names: “It’s quite a number of major pharma companies. The feedback has been very positive. And that’s unfortunately all we can say.”
In addition to helping biopharmas cut their time and expense in developing new drugs, the predictive tox effort holds potential to benefit patients in the developing world, Mundel said in a statement: “Leveraging computation to predict the toxicological risk of drug candidates could ultimately improve productivity across the pharmaceutical industry and unlock major advances against diseases that continue to plague low- and middle-income countries.”
The $10 million grant to Schrödinger is the third grant awarded to the company by the Gates Foundation. Schrödinger received $4,938,764 in 2021 to design and synthesize “highly selective and potent Wee2 inhibitors,” and another $3,495,888 in 2023 to develop Wee2 inhibitors “which could ultimately lead to a new safe and effective nonhormonal contraceptive.”
Last year the Gates Foundation provided $7.7 billion in charitable support, including almost $6.3 billion in grants.
The Gates Foundation’s charitable giving is funded by the BMG Foundation Trust, manager of funds donated by Bill Gates, his former wife and philanthropist Melinda French Gates, and Warren E. Buffett, chairman and CEO of Berkshire Hathaway. As of April 1, the Trust owned 11% of Schrödinger’s common stock (6,981,664 shares) and all 9,164,193 shares of its limited common stock, according to Schrödinger’s proxy statement filed April 25.
Eye on software revenue growth
Should the predictive toxicology solution initiative succeed, Farid said, Schrödinger expects that success to translate to higher software products and services revenues. The company is not publicly predicting when that growth is expected, though it has acknowledged it won’t contribute to revenue growth this year.
Schrödinger saw its software revenues increase 21% year-over-year during the second quarter, to $35.404 million from $29.352 million in Q2 2023. For the first half of 2024, software revenue rose 12% to $68.819 million from $61.565 million in the first three months of 2023.
A successful predictive tox effort is also expected to contribute to the success of future Schrödinger drug development programs. Schrödinger enjoyed a doubling of drug discovery revenues during Q2, to $11.93 million from $5.837 million in April–June 2023, though the company’s six-month drug discovery revenues of $15.113 million lag 61% behind the $38.406 million reported a year ago.
“I think success on this project would absolutely lead to increased demand for the technology, which will of course necessarily translate to growth in the software business,” Farid said. “Our focus is always on building technologies and science that will have an impact. We have a long track record of, when we do that, the demand comes.”
