In the beginning, Genesis Therapeutics spun out from the Stanford University lab of Vijay Pande, PhD, where Evan Feinberg, PhD, then a graduate student, co-invented and co-authored key peer-reviewed papers detailing deep learning technologies.
Notable among them was PotentialNet, the influential neural network algorithm that pioneered the use of novel graph neural networks for molecular property prediction, specifically protein–ligand binding affinity. Feinberg, Pande, and colleagues demonstrated PotentialNet’s performance in potency prediction, further validated through a collaboration between Stanford and Merck Research Laboratories, where Feinberg served as a deep learning consultant before launching Genesis.
Genesis was founded in 2019, and a year later won a $52 million Series A financing. The company has grown since then to raise more than $300 million. Most of that consists of a $200 million Series B financing round completed in 2023. Its investors included NVentures, the venture capital arm of Nvidia, the Silicon Valley-based microprocessing giant that has expanded its market-leading footprint in AI chips to the life industries that include the life sciences.
NVentures recently raised its stake in Genesis by investing what Genesis founder and CEO Feinberg, said was an undisclosed “incremental additional amount” in his company.
Through its collaboration with Nvidia, Genesis is working to accelerate the development of its AI platform, Genesis Exploration of Molecular Space (GEMS). GEMS is designed to generate and optimize molecules for complex targets by integrating proprietary AI methods that include language models, diffusion models, and physical machine learning (ML) simulations.
The additional financing from NVentures is intended to further the capabilities of Genesis’ physical AI platform for structure-driven drug design by applying Nvidia’s expertise to make computation more efficient for several neural network architectures relevant to drug discovery.
“Nvidia is the leader in many aspects of the AI stack, both in terms of hardware, but also the lower-level software layers on top of that hardware, whereas Genesis has been pioneering molecular AI as an intellectual area,” Feinberg told GEN Edge. “So, there’s a lot of very clear synergies between Nvidia’s comparative advantages and Genesis’s comparative advantages that make the combination more than the sum of the parts.”
Optimizing neural networks
The collaboration will, among other areas, encompass optimizing equivariant neural networks, which according to Genesis are valuable for handling 3D geometric data such as protein and small molecule structures.
Nvidia has consistently worked to accelerate computation through neural networks, both training the networks as well as running inference—using trained models to make predictions on new data—or deploying them in a real-world setting.
“For our field of molecular AI that Genesis has been pioneering for years, there are specific types of neural networks that are particularly useful. And that’s actually the continuation of a long trend in the space, where AI is not a monolith. There are many subfields of artificial intelligence that use related but distinct algorithms for learning.”
At Stanford, Feinberg, Pande, and one group of colleagues presented the PotentialNet family of graph convolutions in a 2018 paper in ACS Central Science. Two years later, another group of colleagues joined Feinberg and Pande in showing how, by representing each molecule explicitly as a graph, they achieved “to our knowledge, unprecedented accuracy in prediction of ADMET [absorption, distribution, metabolism, elimination, and toxicity] properties,” showing significant superiority of AI algorithms—a relative 52% and absolute 0.16 increase in R2 versus Random Forests in ADMET prediction—over the advanced ML used by Merck Research Laboratories in a paper published in the Journal of Medicinal Chemistry.
Pande is now general partner at Andreessen Horowitz (a16z) and the founding general partner of a16z’s bio funds, where he leads the firm’s investments that cross biology, computer science, and engineering. Pande, who served as Feinberg’s PhD advisor, led Genesis’ $4.1 million seed round for a16z and co-led for a16z the company’s $200 million-plus Series B, with an undisclosed U.S.-based life-sciences-focused investor, and with Felicis Ventures serving as a major investor in that round.
“I’ve had the really great fortune to be able to work with him for almost a decade at this point,” Feinberg said of Pande. “And I think it’s uncommon to be able to have to work so closely and both learn from and work with someone of that brilliance and vision.”
Evolving with the field
“He [Pande] has just constantly pushed me in a way that’s been really instrumental to the success that Genesis has had. And he continues to just constantly evolve as the field has evolved,” Feinberg added. “I think that’s parallel to our own maintenance of our status as a leader in the field, if that makes sense, and constantly innovating and not just being comfortable emulating, but instead actually pushing the field forward.”
During his graduate student days at Stanford, Feinberg recalled, AI primarily made its impact felt in computer vision and natural language.
“The neural network types that were used for both were actually quite distinct from each other, but neither were very applicable to chemistry. So, we developed new types of neural networks,” Feinberg recalled. “In the mid-2010s, it was graph neural networks that were better suited for molecules.”
Between then and now, Feinberg said, Genesis has consistently worked on new AI algorithms, “new neural network primitives that are better suited for the tasks of molecular AI.”
“Equivariant neural networks is one of those families that is important to us. And that is one of the areas that Nvidia is particularly helping us optimize,” Feinberg added.
Pande’s lab initially rose to prominence through his founding of Folding@Home, the distributed computing project designed to simulate protein dynamics, including the process of protein folding.
“Folding@Home used enormous numbers of Nvidia GPUs [graphics processing units] across the planet to do protein folding simulations,” Feinberg recalled: “Subsequent to that, Nvidia GPUs started to be used much more for artificial intelligence, specifically in vision and natural language. So, we as a company had already been, I would say, power users of Nvidia GPUs.”
“Very natural fit”
“When we were introduced to Nvidia and NVentures through Series B, it felt like a very natural fit for an investor that would not only bring significant capital but also intellect as part of that relationship as well,” Feinberg said. “That investment then formed really the basis to have a relationship that grew beyond being a customer, but actually learning from each other as well, from our needs, and from their lower level capabilities that we could uniquely exploit given our domain knowledge.”
For Nvidia, the collaboration with Genesis bolsters its ongoing efforts to apply AI toward drug discovery.
“Genesis’ AI platform, and related computational advancements developed in collaboration with Nvidia, will help deliver novel generative and predictive AI techniques to explore untapped chemical pathways and identify drug candidates,” said Mohamed “Sid” Siddeek, corporate vice president at Nvidia and head of NVentures.
How will GEMS help Nvidia do both?
“The goal of GEMS is to be able to drug extremely challenging, in some cases, undruggable targets. And in order to do that, we need to accomplish several capabilities better than has existed before,” Feinberg said.
Potency, selectivity, and atomy
That includes generating molecules and predicting their potency, selectivity, and atomy characteristics—a joint, multi-parameter optimization approach to drug discovery for all key characteristics of a molecule together. GEMS consists of two deeply integrated pillars, Feinberg explained—generative AI and predictive AI—and has used Genesis’ own custom language models to generate anywhere from thousands to millions or billions of compounds in the cloud.
“But chemistry, synthetic chemistry is rate limiting. One can only make so many molecules in a given time. So it’s critical that our predictive AI technologies that predict potency, selectivity, and atomy are as accurate as possible. So, GEMS really is an umbrella that describes a deeply integrated set of technologies together,” Feinberg said.
Using GEMS, Genesis is developing a pipeline focused on oncology and immunology. In oncology, Genesis is in late lead optimization phase, approaching the nomination of what it says will be highly potent and selective development candidates for pan-mutant allosteric inhibitors of PIK3CA, an oncogenic driver common to breast and colorectal cancers.
Other oncology development efforts focus on small molecules intended to overcome response to checkpoint inhibitors (lead optimization phase) and prevent evasion of apoptosis in cancer cells by inhibiting an anti-apoptotic regulator of the extrinsic cell death pathway (discovery phase).
In immunology, Genesis says it has two discovery-phase efforts—one to develop multiple programs for generating small molecules aimed at well-validated autoimmune disorder targets; the other, a treatment for “a severe, genetic autoinflammatory disease” using small molecule correctors to restore activity in an unspecified impaired protein.
Collaborations with giants
In addition to its in-house development efforts, Genesis is pursuing collaborations announced with three biopharma giants, about which Feinberg said the company could not comment. The most recent was launched in September with Gilead Sciences, which agreed to discover and develop small molecule therapies across multiple targets, using GEMS to assist in generating and optimizing molecules for Gilead-selected targets.
Gilead agreed to pay $35 million across three targets and holds an option to nominate additional targets for an undisclosed predetermined per-target fee. Gilead also agreed to pay additional payments tied to achieving preclinical, development, regulatory, and commercial milestones, plus tiered royalties on net sales of commercialized products.
The other two collaborations with biopharma giants:
• Eli Lilly—Up-to-$670 million partnership ($20 million of it upfront) to discover novel therapies for up to five targets across a range of therapeutic areas, initiated in 2022.
• Genentech, a Member of the Roche Group—a multi-target, multi-disease effort launched in 2020, using Genesis’ platform for deep learning and molecular simulation. In 2022, Genentech described its targets of interest as “challenging targets that would elude other methods.” The value of the collaboration has not been disclosed.
Genesis is headquartered in the San Francisco suburb of Burlingame, CA, with a fully integrated laboratory in San Diego. The company employs about 80 people.
“We do have a significant amount of expected growth, and that’s partially driven by both the Series B, this latest further investment from Nvidia, and our partnerships,” Feinberg said. “I don’t have a precise number where we’ll be in 12 months, but we do have substantial headcount to grow above that 80.”
The post Nvidia’s Venture Arm Raises Stake in AI Drug Discoverer Genesis Therapeutics appeared first on GEN – Genetic Engineering and Biotechnology News.
