Can all cancers become treatable and metastatic disease be halted in its tracks? Many companies share the goal of ensuring that new therapeutics address early cancers, recalcitrant protein targets, recurrence, and more.
Some approaches nearing fruition are based on decades of foundational research to transform existing platforms to cancer-tracking destroyers such as virus-like particle-drug conjugates, combining the precision of antibody targeting with the catalytic power of protein degradation and creating novel tri-complexes that block oncogenic protein signaling.
Other companies leverage advanced technologies and massive computing power to address difficult-to-treat diseases. Generative AI approaches are being increasingly employed to uncover new targets by taking an unbiased look at the complex biology that drives disease, which basic science and computational skills cannot achieve.
Results are showing promise as many new treatments progress through clinical trials. Although solutions will not arise overnight, companies are determined to treat previously undruggable targets in the hopes of positively impacting the lives of many.
Treating early cancers
Virus-like drug conjugates (VDCs) are being explored as a potentially transformative oncology therapeutic. “We want to shift the paradigm in cancer care, to upfront treatment of the tumor, inducing cytotoxicity and igniting the immune system to recognize and attack tumor-specific neoantigens,” said Jill Hopkins, MD, CMO and president of R&D at Aura Biosciences.
In foundational non-clinical research, over 100 different cell lines and 15 animal tumor models showed that VDCs demonstrate tumor mutation-agnostic activity with high potency.
The targeted cancer therapy, bel-sar (AU-011), a novel VDC, is comprised of a modified human papillomavirus (HPV) virus-like particle (VLP) conjugated with about 200 molecules of a photoactivatable drug that becomes cytotoxic upon activation. AU-011 is made of two recombinantly expressed modified HPV-derived capsid proteins that self-assemble.
1. Kines RC, et al. Int J Cancer. 2016;138(4):901–11. 2. Kines RC, et al. Mol Cancer Ther. 2018;17(2):565–74. 3. Kines RC, et al. Cancer Immunol Res. 2021;9:693–706.
Modified HSPGs (heparan sulfate proteoglycans) are expressed on the surface of cells in the early stages of malignant tumor transformation and do not appear elsewhere. “The VDCs bind to the HSPGs and, with light activation, induce rapid tumor cell necrosis and release tumor neoantigens with local delivery,” said Hopkins. Then, immune cell activation kicks off immune surveillance over time.
The conjugation of dye to VLP with light activation is feasible both from a clinical and manufacturing standpoint. With local delivery, the VDC rapidly reaches the tumor site and binds and infiltrates the tumor cells. Light activation and the release of reactive oxygen species start the necrotic cascade and secondary immune cell infiltration.
Early choroidal melanoma, Aura’s lead indication, is often detected early but, due to treatment toxicity, is not typically treated early. A Phase II trial of bel-sar showed 80% tumor control rate at 12 months (after an initial three months of dosing), and 90% preservation of vision (n=10). The potential treatment is now in a global Phase III trial.
Early NMIBC (Non-Muscle Invasive Bladder Cancer) often presents with blood in the urine. Aura’s Phase I proof-of-concept bladder study of bel-sar showed an absence of tumor cells in the treated tumor in four out of five intermediate-risk patients. In some cases, biopsies from non-treated lesions in the bladder showed a complete response or immune cell infiltration.
“If you could treat a highly recurrent, burdensome, and costly to treat cancer and then induce long-term immune surveillance to impact recurrence… that would be the Holy Grail,” emphasized Hopkins.
Eliminating intracellular targets
Dual-Precision Targeted Protein Degradation (TPD²®) combines a targeted protein degrader payload, such as a PROTAC (proteolysis targeting chimera) or molecular glue, with an antibody for precise, cell-specific delivery.
“By design, degrader–antibody conjugates (DACs) created using our TPD² approach unite the precision of antibody targeting with the catalytic power of protein degradation, enabling the selective elimination of intracellular targets once considered undruggable,” explained Dorin Toader, PhD, head of platform technology at Orum Therapeutics. Conjugation to an antibody improves the degrader’s pharmacological properties and tumor exposure, enhancing both potency and safety.
A key component, the PROTAb linker platform, allows conjugation of a degrader payload to a cell-targeting antibody. This design ensures stability in the bloodstream, keeping the payload inactive until it is released inside the target cell in its native form, while enabling receptor-mediated endocytosis and precise intracellular delivery.
“While PROTAb represents a major advance in targeted delivery, we view it as a foundation for continued innovation rather than a complete solution to every delivery challenge,” clarified Toader.
The company has developed a payload targeting GSPT1, a key regulatory protein involved in cancer cell survival. Screenings across more than 250 cancer cell lines identified several highly sensitive tumor types, including acute myeloid leukemia (AML), small cell lung cancer (SCLC), and neuroendocrine tumors (NETs). Orum programs include ORM-1153 in preclinical development for AML and other hematologic malignancies, and ORM-1023 in pre-candidate stage for SCLC and NETs. Also of interest is the discovery of novel degrader payloads that target additional E3 ligases and proteins of interest.
The company is advancing both its internal pipeline and its collaboration with Vertex Pharmaceuticals, while BMS continues to progress the ORM-6151 program it acquired (now named BMS-986497).
The field of targeted protein degradation is advancing rapidly and continues to redefine how to address difficult disease targets. “We expect to see a growing number of proteins once considered undruggable successfully degraded, broadening the therapeutic reach of this modality beyond oncology into neurodegenerative, inflammatory, and infectious diseases,” said Toader.
Advances in AI, machine learning, and structural biology are also helping to accelerate the rational design of degraders and expand the range of usable E3 ligases.
Inhibiting common drivers
The discovery of small-molecule inhibitors requires identifying suitable binding pockets on protein surfaces. Proteins lacking such pockets require innovative strategies for therapeutic targeting, such as the use of machine learning and AI in drug discovery efforts.
Oncogenic RAS mutations are among the most common drivers of human cancer often associated with poor survival outcomes, including approximately 90% of pancreatic ductal adenocarcinoma (PDAC), 50% of colorectal cancer and 30% of non-small cell lung cancer (NSCLC).
Normal RAS proteins function as molecular switches that cycle between “ON” and “OFF” states, with “RAS(ON)” signaling within a cell directing its proliferation. Excessive RAS(ON) signaling caused by RAS mutations drives much of cancer initiation and progression. These mutant RAS proteins have long been recalcitrant drug targets.
“Our RAS(ON) inhibitors are designed to suppress diverse oncogenic variants of RAS proteins,” said Mark A. Goldsmith, MD, PhD, CEO and Chairman of Revolution Medicines. “Using structure-guided design, we created molecules that bind to the chaperone protein cyclophilin A to form a novel tri-complex that blocks oncogenic RAS signaling.”
“Our research collaboration with Iambic Therapeutics builds on the progress of our RAS(ON) inhibitor programs,” continued Goldsmith. “The combination of their AI-driven discovery technology with our proprietary data and drug discovery capabilities presents an opportunity to rapidly explore challenging oncology targets.”
Granted FDA Breakthrough Therapy Designation in previously treated metastatic pancreatic cancer with KRAS G12 mutations, Daraxonrasib, a RAS(ON) multi-selective inhibitor, is active against diverse RAS driver mutations and multiple drug resistance mechanisms. Two Phase III trials are ongoing—one in patients with second-line metastatic PDAC and one in patients with metastatic previously-treated NSCLC. In addition, a Phase III trial of first-line treatment in patients with metastatic PDAC will be initiated.
Elironrasib, a RAS(ON) G12C-inhibitor, binds selectively and covalently to the oncogenic RAS(ON) form of the RAS G12C variant that drives about 12% NSCLC cases. Elironrasib also was granted FDA Breakthrough Therapy Designation for KRAS G12C-mutated locally advanced or metastatic NSCLC patients who have received prior treatment but not been treated with a KRAS G12C inhibitor.
The G12D variant is the most common RAS mutant that causes solid tumors. Zoldonrasib, a RAS(ON) G12D-selective covalent inhibitor, demonstrates acceptable safety and tolerability, and shows encouraging initial antitumor activity in NSCLC patients.
Improving pharmaceutical research
Pharma.AI, a commercially available, end-to-end, generative AI software platform designed to improve the quality and productivity of pharmaceutical research. Used by 13 of the world’s top 20 pharmas, the AI software spans biology, chemistry, clinical development, and science research.
AI-powered target discovery leverages advanced generative models to evaluate complex omics and clinical datasets to identify novel, commercially viable, high-confidence targets, even those previously considered intractable. Deep learning, causality inference, and literature mining provide novelty, tractability, and scientific confidence. Next, the best possible molecules are identified that satisfy the rules of classical medicinal chemistry.
“Generative AI allows you to design perfect diamonds, so to speak, already with proper polishing, from scratch,” said Alex Zhavoronkov, PhD, Founder and CEO of Insilico. “It can achieve multi-parameter optimization for 30 different properties.”
In the early development of Rentosertib for idiopathic pulmonary fibrosis, generative AI was central at every stage. Trained on relevant omics and clinical datasets, the AI platform PandaOmics pinpointed TNIK, a protein previously unrecognized as a fibrosis target.
Then the generative chemistry engine, Chemistry42, was used to design and optimize novel molecular structures through structure-based drug design and iterative feedback. “Within several rounds, we discovered a potent lead compound with a nanomolar IC50, and further optimized it for drug-like properties. Remarkably, the lead molecule, Rentosertib, was identified after synthesizing and testing fewer than 80 compounds,” said Zhavoronkov.
Preclinical studies confirmed promising efficacy and safety. The entire process—from initial target prediction through preclinical candidate nomination—was completed in less than 18 months. Rentosertib has completed a Phase IIa clinical trial, the industry’s first proof-of-concept validation for AI-driven drug discovery.
Since 2021, Insilico has built a pipeline of 31 programs for 29 drug targets in areas of high unmet need, including fibrosis, oncology, immunology, and other areas. ISM5411, a PHD1/2 inhibitor for inflammatory bowel disease, has completed Phase I clinical trials and three anti-tumor programs have entered the initial patient dosing phase.
“The true test lies in practical, real-world application,” stated Zhavoronkov. “An ongoing project is building ‘pharmaceutical superintelligence,’ a fully autonomous platform capable of discovering and designing optimal small molecules or biologic drugs, alongside robust biomarkers.”
Unbiased view of disease
Traditional approaches often focus on a single, well-known target. But biology is far more complex. The Recursion OS (Operating System) specifically helps address complex problems by discovering novel biology and enabling precision, AI-driven chemistry.
“Through our maps of biology, we are able to move beyond a reductionist model toward a holistic, unbiased view of disease,” said Najat Khan, PhD, Chief R&D Officer and CCO at Recursion. “We combine massive data generation—millions of experiments weekly—with AI models that can predict trillions of gene-compound relationships to identify new pathways and mechanisms to engage difficult targets.”
These complementary approaches, including generative AI, active learning, and others, are delivering a portfolio of potentially first- and best-in-class medicines for cancer. “In our partnership with Bayer, we’re focused on precision oncology programs, with Roche and Genentech on neuroscience and GI oncology, and with Sanofi on oncology and immunology,” detailed Khan.
The company’s internal therapeutic pipeline includes the REC-1245 program, an example of identifying an alternative approach. Historically, CDK12 has been extremely difficult to target without impacting the closely related CDK13, which can cause serious dose-limiting toxicities.
Using AI-powered maps of biology and the Recursion OS platform, Recursion identified that RBM39 appeared to be functionally similar to CDK12. An RBM39 degrader was designed (REC-1245) that mimicked the response of genetically knocking out CDK12, without impacting CDK13.
AI (ClinTech) was leveraged to inform a biomarker-enriched clinical development strategy. REC-1245 is now in a Phase I/II study with enrollment focused on patients with biomarker-selected solid tumors and lymphoma.
Other programs in the clinic include REC-617, an orally bioavailable, highly potent, and selective CDK7 inhibitor for advanced solid tumors; REC-3565 a small molecule MALT1 inhibitor for multiple hematology indications; and REC-4881 an orally bioavailable, allosteric small molecule inhibitor of MEK1 and MEK2 to reduce polyp burden and progression to adenocarcinoma in people living with familial adenomatous polyposis (FAP), a rare tumor predisposition syndrome.
“We’re just scratching the surface of the promise of AI in both biology and chemistry,” said Khan. “The immediate challenge is about integration, bringing together all of the individual tools into a single, unified, end-to-end platform. We need to ensure that every single AI investment is evidence-based, technically feasible, and directly tied to delivering real, measurable value for patients and the pipeline.” design, and clinical development to identify new pathways and mechanisms to engage difficult targets, design identified candidate molecules, and inform biomarker-enriched clinical development strategies.