NEW YORK – Insilico Medicine recently advanced into clinical trials a drug it designed and discovered using generative AI, the same technology that powers OpenAI's ChatGPT and Midjourney.
The algorithms that underlie these tools are built using a type of machine learning called a generative adversarial network, in which dueling neural networks compete using deep learning to hone the accuracy of predictions made in response to a text, image, or other prompt. While ChatGPT and Midjourney have made headlines by generating natural language text, synthetic images, and algorithms, generative AI can similarly be applied to discover drugs.
Insilico has leveraged its suite of AI-based tools to advance a pipeline of drug candidates, including a clinical stage USP1 inhibitor, ISM3091, for treating BRCA1/2-mutant solid tumors, and preclinical candidates for MTAP-deleted cancers and estrogen receptor-positive, HER2-negative breast cancer.
Insilico, founded in 2014 by CEO Alex Zhavoronkov, was one of the first groups to publish a method that uses an adversarial model for new compound generation. The Hong Kong- and New York-based company accomplished that by developing deep learning architectures that combine generative algorithms with reinforcement learning and applying them to chemistry and pharmacology data to produce novel molecular structures with predefined properties.
Insilico's Pharma.AI suite of tools includes PandaOmics for discovery and prioritization of novel targets; Chemistry42 for generating new small molecule drug candidates; and InClinico for predicting clinical trial success rates. Using the platform, Insilico has advanced 31 programs for 29 drug targets, including eight oncology programs.
"One of the characteristic features of PandaOmics is its knowledge graph, or target disease associations, which is generated by transformer-based natural language processing models applied to biomedical information published in the scientific literature over the last 30 years," said Insilico Chief Medical Officer Sujata Rao.
This algorithm can be used to understand and generate human language and extract information from sources to create a framework that aids in target discovery. In March, Insilico introduced ChatPandaGPT, an advanced AI chat functionality integrated into the PandaOmics knowledge graph, to expedite information gathering related to molecular biology, target discovery, and drug development. Rao said using ChatPandaGPT, biologists and clinicians can broadly interpret biological data through natural language conversations with the platform. "It's basically focusing on storytelling in data analysis and providing you guidance at each step," Rao said.
Chemistry42, launched in 2020, integrates AI algorithms with computational and medicinal chemistry methods to generate novel molecules with drug-like properties. Rao said it uses both structure-based and ligand-based drug design and integrates with PandaOmics to allow a seamless workflow from novel target identification to de novo small molecule generation and optimization.
Rounding out Insilico's suite of tools, InClinico is a generative AI-powered software platform launched in November 2022. It uses data including small molecule chemical properties, transcriptomics, literature, and clinical trial protocols to predict the outcome of Phase II clinical trials. In a July 2022 preprint publication, company researchers reported the platform predicted that 17 out of 35 Phase II trials conducted by various pharma companies will meet their primary endpoints. Once all these trials read out, the real-world success rates will be compared to InClinico's predictions and provide insight into the platform's performance.
Insilico's most advanced cancer drug program, the USP1 inhibitor ISM3091, targets upregulated deubiquitinase in tumors with BRCA1 mutations. Rao explained that inhibition of USP1 results in persistence of mono-ubiquitinated PCNA at the replication fork, resulting in replication fork destabilization, which leads to decreased viability of BRCA1-deficient cells, causing a synthetic lethal effect.
Two PARP inhibitors, Merck and AstraZeneca's Lynparza (olaparib) and Pfizer's Talzenna (talazoparib) are approved in the US for breast cancers with BRCA1/2 mutations. However, not all patients respond and even among those who do, resistance develops. "We know there's a critical need for agents that can overcome primary and acquired PARP inhibitor resistance," Rao said. "Inhibition of USP1 may be a useful treatment for [BRCA1-mutated] tumors with acquired replication fork stabilization."
Insilico presented data at the American Association for Cancer Research annual meeting in April, showing that ISM3091 induced a dose-dependent increase in PCNA ubiquitination in triple-negative BRCA1-mutated breast cancer cell lines and inhibited growth of BRCA1-mutated cells in vitro and in vivo. ISM3091 further demonstrated dose-dependent monotherapy activity in BRCA1-mutated triple-negative breast cancer cell line-derived xenograft models, and the combination of ISM3091 and Lynparza showed promising synergistic effects in vivo. ISM3091 also had promising activity in BRCA1-mutated ovarian cancer, in BRCA1/2 wild-type lung cancer models, and in a PARP inhibitor-resistant patient-derived xenograft model.
In an upcoming Phase I trial, Insilico plans to study ISM3091 in BRCA1-mutated tumors either alone or in combination with a PARP inhibitor, chemotherapy, or other investigational drugs. Researchers will also assess biomarkers of response such as ubiquitinated PCNA. The trial will be conducted in the US and China simultaneously, and Insilico expects to begin the study at the first US center in July.
Another one of Insilico's pipeline candidates, ISM5939, an inhibitor of ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1), is entering investigational new drug application-enabling studies. ENPP1 is involved in regulating immune, cardiovascular, neurological, and hematological system functions. In certain cancers, elevated ENPP1 expression has been linked with metastasis and poor prognosis, and inhibition of ENPP1 has been shown to enhance the anti-tumor effects of the immune system.
Insilico used its Chemistry42 generative AI platform to design ISM5939 with a structure that is significantly different from that of other ENPP1 inhibitors. According to the firm, the drug has shown robust anti-tumor efficacy in vivo and a favorable safety, toxicity, and pharmacokinetic profile in vitro. Supriya Bavadekar, Insilico's senior medical director, said the company has not yet decided on a patient selection strategy for ISM5939 when it enters clinical trials. Because ENPP1 inhibition seems to promote remineralization of bone, Insilico is also studying the drug in hypophosphatasia, a rare genetic disorder that affects development of the bones and teeth.
According to Rao, commercial strategies for ISM3091 and ISM5939 are "evolving," but with the advancement of these two drug candidates toward clinical studies, the company is pursuing partnerships and out-licensing opportunities.