The rapid growth and spread of cancer cells in the body can be caused by gain-of-function mutations in oncogenes, resulting in the increased ability of such cells to proliferate. A common drug discovery strategy is to inhibit overactive or overexpressed proteins that are produced as a result of the oncogene mutation activities, with a number of approved drugs for many cancer types. However, more than half of all known cancer mutations lead to loss of function in another group of genes — the so-called tumor suppressor genes — natural regulators that prevent uncontrolled cell growth. These mutations, along with other mechanisms for functional loss such as deletions and expression silencing, are much harder to target. It is challenging to modulate the activity of a faulty protein or replace a protein that isn't there.

One strategy to go after loss-of-function mutations is to exploit synthetic lethality, a promising pathway to develop anti-cancer treatments that will benefit a whole new set of cancer patients.

Synthetic lethality refers to a situation where simultaneous alterations in two different genes result in cell death, but a mutation in either gene alone does not. The word synthetic is used here for its ancient Greek meaning: the combination of two entities to form something new.

The phenomenon of synthetic lethality was first described by the American geneticist Calvin Bridges in the early 20th century, when he noticed that some combinations of mutations in the model organism Drosophila melanogaster (the common fruit fly) rendered lethality. When crossing two fruit flies, certain non-allelic genes were lethal only in combination even though the homozygous parents were perfectly viable. The term "synthetic lethality" was coined two decades later by his colleague Theodore Dobzhansky who observed the same phenomenon in wildtype populations of another fruit fly species, Drosophila pseudoobscura.

Synthetic lethality is a consequence of the tendency of organisms to maintain "backup plans" for all key processes in cells -- to protect phenotypic stability even when unpredictable events occur, such as genetic variations, environmental changes, or random mutations. Nature maintains genetic robustness by means of having parallel redundant pathways and "capacitor" proteins that camouflage the effects of mutations so that important cellular processes remain unaffected by any individual component. In practical terms, in order to exploit the phenomenon of synthetic lethality one needs to identify these buffering relationships, and understand what happens when they break down. Scientists do this through the identification of gene interactions that function in either the same biochemical processes or pathways that appear to be unrelated.

In the context of oncology drug discovery, the synthetic lethality approach indirectly targets loss-of-function mutations, often in tumor suppressing genes. This approach requires the availability of gene pairs that, when simultaneously inactivated, render death of cancer cells. When such a situation occurs (e.g. in Homologous recombination deficiency [HRD] tumors), the strategy is to then target a second gene, one that compensates for the loss of activity of the dysfunctional gene. The best-known examples in cancer involve the tumor-suppressing BRCA genes, which play a vital role in repairing DNA damage, and PARP genes that act on a "back-up" repair pathway. In this case, cancer cells can tolerate breaking down one pathway or the other, but not both at the same time.

Loss-of-function mutations in BRCA1 and BRCA2 genes are found in roughly 30-40% of familial breast cancer patients and in up to 80% of hereditary ovarian cancer patients, making the synthetic lethality a principle area for anti-tumor drug discovery with a potentially large group of patients who might benefit from such therapeutics. In fact, the first drug exploiting the synthetic lethality phenomenon, PARP inhibitor for treating patients with mutations in BRCA1/BRCA2 genes, was approved in Europe and the United States back in 2014, and since then, three other approved PARP inhibitors followed.

Following initial successes with PARP inhibitors, scientists have searched for new synthetic lethal interactions -- using advances in functional genomic screening and next generation sequencing (NGS) to screen thousands of gene pairs at once. For instance, researchers can use RNA interference or CRISPR gene editing on cells bearing a mutation of interest to systematically disrupt thousands of other genes to try to find a co-dependency that might render cancer cell death.

One promising synthetic lethality target identified as a result of such efforts is Ubiquitin-specific protease 1 (USP1), a 785 amino-acid deubiquitinating enzyme that contains the characteristic His and Cys domains highly conserved in all members of the ubiquitin-specific processing family of proteases.

Ubiquitination refers to the post-translational modification of a protein by the covalent attachment of one or more ubiquitin monomers. The most prominent function of ubiquitin is labeling proteins for proteasomal degradation. Besides this function, ubiquitination also controls the stability, function, and intracellular localization of a wide variety of proteins.

For instance, USP1 was found to regulate the ubiquitination status of Fanconi anemia protein FANCD2, a critical component of DNA damage repair. While DNA-dependent mono-ubiquitination of FANCD2 facilitates DNA repair, USP1 plays a critical role by deubiquitinating the protein to extinguish the timely DNA-repair response. There is also evidence that USP1 plays an important role in DNA translesion synthesis by regulating mono-ubiquitination of PCNA and promotes homologous recombination to repair DNA double strand breaks. Moreover, USP1-deficient mice were observed to be highly sensitive to DNA damage. Interestingly, USP1 is significantly overexpressed in a number of cancers, and blocking USP1 in such cases led to the inhibition of DNA repair and a consequent apoptosis in multiple myeloma cells as well as sensitization of lung cancer cells to cisplatin. These results indicate that USP1 is a promising target for chemotherapy for at least some cancers, for instance, in HRDtumors.

Insilico Medicine initiated an AI-designed program targeting USP1 in June 2021, and today we are happy to announce that we have completed the experimental validation of AI-designed molecules with the desired properties and nominated a preclinical candidate, ISM0003091, for anti-cancer therapies targeting USP1.

Our candidate compound is an orally available selective small molecule inhibitor of USP1, which plays an essential role in the cellular response to DNA damage by deubiquitinating specific proteins in the Fanconi Anemia (e.g., FANCD2) and Translesion Synthesis pathways (e.g., PCNA). In HRDtumors, USP1 inhibition could induce cell cycle arrest and cell apoptosis. The synthetic lethality between USP1 and HRD is due to persistent monoubiquitinated PCNA/FANCD2.

Insilico Medicine's preclinical candidate molecule demonstrated significant potential when targeted against a broad range of tumor lineages with HRD backgrounds. In vitro data showed potent anti-proliferation activity of the compound in BRCA mutant tumor cells with excellent selectivity. In vivo efficacy studies in CDX and PDX models showed strong anti-tumor activity and robust/durable tumor regression both in monotherapy group and when combined with Olaparib. Insilico Medicine has initiated an Investigational New Drug (IND)-enabling study of the compound to advance this internally developed program to the clinical stage.

The novel structure of the compound ISM0003091 was designed and optimized through Insilico Medicine's proprietary AI platform Chemistry42, which utilizes generative adversarial network (GAN) technology to advance small molecule generation.

Using an ensemble of generative and scoring engines that can "imagine" molecules from scratch using cutting-edge deep learning technologies pioneered by Insilico Medicine for pharmaceutical research applications as early as 2015, Chemistry42 automatically creates drug-like molecular structures with appropriate physicochemical properties.

The generative procedure in Chemistry42 engages an asynchronous ensemble of proprietary generative models of different nature. These are carefully selected and curated algorithms with diverse architectures implementing distinct strategies. The platform utilizes multiple machine learning models and molecular representations for different scenarios to maximize their contribution and the platform's efficiency.

For example, some of the models focus on the exploration of the chemical space, while others are tailored to improve these explored structures. In the current version of Chemistry42, there are over 30 generative models, including generative autoencoders, generative adversarial networks, evolutionary algorithms, language models and others. Moreover, these models employ different molecular representations — string-based, graph-based, and 3D-based. It is essential to understand and stimulate the interplay of multiple models. Rather than treating these algorithms as black-box solutions, Insilico Medicine provides deep, domain-specific analytics to understand the advantages and drawbacks of each approach.

Combining various state-of-the-art machine learning methods, Chemistry42 is the most comprehensive generative chemistry effort to date, capable of quickly delivering diverse, high-quality molecular structures in a fast, fully automated manner. As the structures are generated, they are dynamically assessed using the reward and scoring functions of the platform.

The synergy between the capabilities of the Chemistry42 platform and strong internal drug discovery expertise allowed us to achieve nomination of the preclinical candidate in just nine months with less than 80 compounds synthesized and tested since project initiation.

The rapid nomination of ISM0003091 as the preclinical candidate for anti-cancer therapies targeting USP1 in just 9 months is another important milestone in a series of successes at Insilico Medicine, enabled by our end-to-end drug discovery engine pharma.ai. The platform now includes three key components: a target discovery and multi-omics data analysis engine PandaOmics, a de novo molecular design engine Chemistry42, and a clinical trial outcomes prediction engine InClinico.

In the past 12 months our R&D team, enabled by pharma.ai, managed to successfully deliver six preclinical candidates for our internal pipeline, in collaboration with our partners. Some of the latest examples included a novel drug candidate for a novel wide-indication target to treat Idiopathic Pulmonary Fibrosis (IPF), developed in under 18 months and at roughly 1/10th of the typical cost associated with similar programs. Just six months later, we nominated a novel drug candidate for Kidney Fibrosis for our internal drug discovery pipeline. The latest example of a major partnership includes a novel drug candidate for the QPCTL immuno-oncology target, developed using our pharma.ai platform and nominated in under 40 days since the announcement of strategic partnership with Fosun Pharmaceuticals (stock code: 600196.SH, 02196.HK), a leading innovation-driven international healthcare group in China.

One of our AI-discovered molecules, an anti-fibrotic drug candidate, has successfully completed the Phase 0 microdose trial and entered Phase I clinical trial, marking our first internally-developed program for fibrosis.

The application of our proprietary AI-driven engine pharma.ai allowed for all of our drug candidate nominations to be achieved in record time, and at a fraction of the cost of a "traditional" drug discovery workflow. More importantly, the application of cutting-edge AI to the drug discovery process permits us to innovate faster, discovering novel targets, novel mechanisms of action, and potentially first-in-class molecules for indications with highly unmet needs.