[ad_1]
In developing drugs using a platform that joins physics with machine learning, Schrödinger sees more than a passing resemblance to the studio whose Toy Story and other computer-generated movies changed the practice of animation.
Just as Pixar initially viewed its competitors as studios that produced animation using yesteryear’s hand-drawn approach, so Schrödinger pinpoints its rivals not as other users of artificial intelligence (AI) in drug discovery, but chemists who still use traditional methods for designing molecules.
“Our Toy Story in some sense is Nimbus [Therapeutics], and Morphic [Therapeutic], and now our own programs,” Schrödinger CEO Ramy Farid, PhD, analogized to GEN Edge.
Farid was referring to a pair of drug developers whose discovery platforms add computation to traditional chemistry—and which have both partnered with Schrödinger in designing new drugs. Nimbus’ structure-based engine combines computational chemistry with technologies that include X-ray crystallography and cryo-electron microscopy. Morphic integrates its MInT (Morphic Integrin Technology) platform with Schrödinger’s computational prowess to modulate the activity of integrins to treat disease.
Where Schrödinger stands out in drug discovery, Farid said, is in combining AI with physics-based first principles to identify new drugs for targets that are designed to treat a variety of diseases. Applying physics means running a molecular dynamics simulation to compute the solubility of a molecule in water, or the affinity of the molecule for a particular protein, or its permeability.
If physics can do all that and more, why does Schrödinger rely on machine learning too?
“The calculations are slow, relatively speaking. It takes about a day to compute one property on one processor, approximately 12–24 hours. And to do drug discovery, we need to explore hundreds of millions—billions, actually—of molecules,” Farid explained. “Even if you had one million computers, you couldn’t do that—and we don’t have access to one million computers! So, we need this hack, if you will, to generate training sets, with physics that’s pretty fast to generate a large enough amount of data to train a machine-learned model. Machine-learned models are really fast, but you need the data to train them.”
Schrödinger has been using forms of AI stretching back about two decades, to a precursor technology called Quantitative structure-activity relationship, or QSAR, which studied the relationship between the chemical structure (or structure-related properties) and the biological activity of a chemical. As the underlying technology improved, its name changed to deep learning, then neural networks, and now AI.
“With QSAR, you couldn’t handle as much data. And you couldn’t give it as many descriptors because the method just couldn’t handle it,” Farid recalled. “So there have been advances, but the fundamental limitation hasn’t changed. You cannot build a training set in chemistry large enough because of the context. You have 1060 different molecules. You have 30,000 different proteins, and among each of those different proteins, you have different conformations and so on. The complexity is way, way too high for machine learning alone.”
Prongs and pipeline
Farid and Karen Akinsanya, PhD, Schrödinger’s president of R&D, therapeutics, discussed the company’s approach to drug discovery, its successes, and its challenges during a recent interview with GEN Edge at the company’s New York headquarters. Schrödinger is named for a Nobel laureate honored for pioneering discoveries in quantum physics, Erwin Schrödinger, so it stands to reason that the company’s drug discovery efforts are underpinned by physics as well as machine learning.
Akinsanya and Farid said Schrödinger’s success in drug discovery stems from a two-prong approach to its business. One prong is the licensing of its software used in drug discovery and materials design, a business that has attracted some 1,750 customers to the company.
The other prong, which has two components, focuses on full-scale deployment of the platform to drug discovery: The company 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’s partners include several big pharma giants:
- Bristol Myers Squibb, which in 2020 agreed to pay Schrödinger up to $2.75 billion ($55 million upfront, the rest in milestone payments and royalties) toward a collaboration to discover, develop, and commercialize treatments against targets in oncology and neurological disorders. The companies have confirmed work on a preclinical program fighting KRAS-driven cancers, as well as discovery-phase programs against undisclosed targets in immunology and oncology.
- Eli Lilly, which has committed up to $425 million toward a program for an undisclosed indication announced last October.
- Takeda Pharmaceutical, which in 2017 agreed to pay Schrödinger up to $170 million per program under a multi-target research collaboration directed to diseases that align with Takeda’s core therapeutic areas of interest including oncology.
Among smaller biotech collaborators have been Structure Therapeutics, for which Schrödinger played a role in helping to design GSBR-1290, an oral GLP-1R in development for type 2 diabetes and obesity; Relay Therapeutics, whose Dynamo Platform™ joins computation with physics-based simulations and chemical biology insights to achieve what it calls Motion-Based Drug Design®; and Morphic, for which the company has worked in collaboration to design MORF-057, an oral ɑ4ꞵ7 inhibitor in development for ulcerative colitis and Crohn’s disease.
Earlier this month, Schrödinger acknowledged the end of a collaboration with Zai Lab to discover and develop a novel oncology program targeting DNA damage response; Zai Lab cited strategic reasons. (The program, which had been in the discovery phase, is now fully owned by Schrödinger.)
Schrödinger has also built a wholly-owned pipeline of 19 active programs, the first of which entered clinical trials last year—SGR-1505, an inhibitor of the mucosa-associated lymphoid tissue lymphoma translocation 1 gene (MALT1), now in two Phase I trials. SGR-1505 is under study in Australia in healthy adult volunteers (ACTRN12623000358640p), with preliminary data expected later this year. Another trial (NCT05544019) is assessing SGR-1505 in patients with relapsed/refractory B-cell lymphomas. The company expects to have preliminary data for the latter study next year, Akinsanya said.
Comparing PK/PD profiles
Michael J. Yee, equity analyst with Jefferies, observed recently in a research note that Schrödinger executives expressed confidence that SGR-1505 can avoid the off-target toxicities seen with Johnson & Johnson’s MALT1 inhibitor candidate, safimaltib (JNJ-67856633), because of what Schrödinger said was a more attractive pharmacokinetics and pharmacodynamics (PK/PD) profile.
Yee noted that safimaltib has shown dose-limiting toxicities (DLTs) as well as promising efficacy. “[Management] thinks the DLTs [dose-limiting toxicities] with the JNJ drug may be due to the property of the molecule (not the target) since it has 13x variability in PK between [patients] and a long 5-day half-life which requires loading doses even at high doses, leading to very high drug exposure,” Yee wrote.
“SDGR sees opportunities in [monotherapy] as well as combo with BTK [Bruton’s tyrosine kinase] and/or other targets, and is open to partnering with companies with these assets or presence in B-cell tumors.”
Also seeing potential in a combination therapy is J&J. A study published in February showed effectiveness in treating mantle cell lymphoma through a combination therapy of safimaltib and Eli Lilly’s Jaypirca™ (pirtobrutinib), a BTK inhibitor granted accelerated approval by the FDA in January as the first and only non-covalent (reversible) BTK inhibitor authorized by the agency.
Yee has been bullish on Schrödinger. In June, he raised his firm’s 12-month price target on Schrödinger stock 50%, from $40 to $60. “SDGR is a rare model in biotech overlapping the AI + ML theme w/ both a strong commercial software biz (beat last two years guidance) + their own internal cancer pipeline in Phase I and partnerships across pharma/biotech demonstrating increasing proof of concept.”
Indeed Schrödinger has disclosed that it received $111 million upon completion in February of Nimbus’ sale of its wholly-owned Nimbus Lakshmi subsidiary and its tyrosine kinase 2 (TYK2) inhibitor NDI-034858 to Takeda for up to $6 billion (of which $4 billion was paid upfront). Schrödinger also received another $35.79 million from Nimbus in April.
Last month, Nimbus and Schrödinger published a paper highlighting the impact of computational physics-based predictions in the identification of NDI-034858—since renamed TAK-279, which is in Phase II clinical trials for the treatment of psoriasis and psoriatic arthritis.
Second, third clinical candidates
Also later this year, a second wholly-owned Schrödinger candidate is expected to advance into the clinic: SGR-2921, an inhibitor of the cell division cycle 7 (CDC7) gene that is a key cell cycle checkpoint for DNA repair. SGR-2921 has received FDA clearance for a Phase I study (NCT05961839) in patients with acute myeloid leukemia or myelodysplastic syndrome. “We are initiating the trial right now,” Akinsanya said.
Next year, Schrödinger expects to bring a third candidate into the clinic, SGR-3515, a Wee1 inhibitor indicated to treat gynecological cancers with additional potential in a broad range of solid tumors. SGR-3515 is now in IND-enabling studies.
Schrödinger plans additional updates on these and other candidates in December when it updates analysts and investors on its pipeline progress. “We anticipate that our team will be moving programs into the clinic and through Phase I over the course of the next decade,” Akinsanya said.
Facilitating that progression of candidates among the wholly owned pipeline is among the clinical development and regulatory strategy responsibilities of Schrödinger’s recently appointed chief medical officer, Margaret Dugan, MD.
A board-certified medical oncologist and hematologist with more than 30 years of clinical, medical research, and drug development experience, Dugan was previously chief medical officer of Dracen Pharmaceuticals, an oncology drug developer, and held roles of increasing responsibility, including senior vice president, at Novartis, where she led oncology-focused global strategic drug development.
Oncology is one of three therapeutic areas that account for most of Schrödinger’s wholly-owned pipeline; the other two are neurology and immunology. “This is a neat consequence of our physics-based methods: They are first principles methods, so they are completely agnostic to the target class, the therapeutic area,” Farid said. “It’s not just focusing on a particular target or target class or therapeutic area. But of course, it’s necessary to focus somewhere.”
Added Akinsanya: “We really focus on the types of targets that we think are amenable to our platform, and that have what we call strong human evidence.”
“We don’t think it makes sense for Schrodinger to pursue very speculative biology in Nature papers. We are influenced a lot by very interesting drug design challenges, where there’s either clinical data or even an approved drug, where there’s an opportunity for a different kind of molecule, or where there’s an opportunity to be first in class on a highly exciting but validated from a genetics point of view target.”
Quarterly review
Schrödinger finished the second quarter with net income of $4.3 million, improved year over year (YOY) from a net loss of $47.7 million in the second quarter of 2022. Revenue, however, fell nearly 9% YOY from $38.5 million to $35.2 million.
The revenue dip largely reflects a 32% drop in drug discovery revenue, from $8.5 million in Q2 2022 to $5.8 million during April–June. Schrödinger has since lowered its guidance to investors on the drug discovery revenue it anticipates generating for all of 2023, from a range of $70–$90 million, to a range of $50–$70 million. The company finished last year with $45.4 million in drug discovery revenue, up 84% from $24.7 million in 2021.
Drug discovery revenue hinges on the timing of milestones achieved by collaboration partners, with delays largely accounting for the drop, Farid said. “Once we deliver a development candidate in the cases where we’re doing that, then it’s in the partner’s hands, and they are running the clinical trials. To the extent that some milestones are associated with clinical trials, again it’s beyond our control,” Farid added. “That means it’s really hard to predict in what quarter certain milestones are going to come in.”
Also declining YOY was software revenue, which dipped 2% to $29.4 million from an even $30 million in the second quarter of last year. But the quarterly result still placed it squarely within the company’s $27-31 million guidance range for Q2, which Schrödinger is maintaining for Q3.
“Here’s the challenge: The software business is seasonal because the revenue has to be recognized in the quarter that the customer paid for licensing the software. So what we have is most of our customers have a license that starts in Q4 or Q1,” Farid said. “That’s why Q4 and Q1 look like these really big quarters and Q2 and Q3 are smaller quarters. So, the software business is on track.”
Schrödinger has raised its 2023 investor guidance on software revenue, by projecting an overall 15%-18% increase this year, up from a range of 13–17%. The company racked up $135.6 million in software revenue last year, up 20% from $113.2 million in 2021.
A software guidance increase, according to Yee, “is rare as they never raise guidance because Q4 is the big quarter” when many customers agree to license Schrödinger’s software: “This implies visibility; things are percolating for Q4.”
[ad_2]
Source link