AI builds dual-action cancer drug targeting PKMYT1

Generative artificial intelligence (GenAI) is increasingly being applied across early drug discovery, from target identification to molecular design. One emerging application is the development of targeted protein degraders – molecules designed to remove disease-associated proteins. Many degraders are bifunctional, binding both the target protein and proteins involved in cellular degradation, creating significant structural and optimisation challenges that computational approaches can help address.

Recent research published in Nature Communications shows how Insilico Medicine combined generative AI and structure-guided design to develop a dual-action drug candidate targeting PKMYT1, a kinase implicated in tumour growth. The molecule is a targeted protein degrader that both inhibits PKMYT1 activity and eliminates the protein from cells. The study provides an end-to-end example of AI-assisted design, from identifying a novel inhibitor chemotype to developing a bifunctional molecule with both inhibitory and degradative activity.

I helped lead the effort to use generative AI and structure-guided design to create a dual-action degrader targeting PKMYT1

The research was led in part by Dr Alex Zhavoronkov, founder and CEO of Insilico Medicine, whose work focuses on applying AI to accelerate drug discovery and development.

“I helped lead the effort to use generative AI and structure-guided design to create a dual-action degrader targeting PKMYT1,” Zhavoronkov states.

Insilico Medicine has focused on AI-driven drug discovery since its founding in 2014. Initially the company concentrated on developing deep learning approaches for biomedical applications before expanding into therapeutic programmes in immunology, oncology, fibrosis and central nervous system disorders. Since 2019, the organisation has nominated multiple preclinical candidates and advanced several molecules into clinical development, including programmes where both the therapeutic target and the molecule were identified using AI.

PKMYT1 as a drug discovery target

PKMYT1 is a cell cycle regulatory kinase that has emerged as a potential oncology target. Genetic and functional studies suggest that inhibiting PKMYT1 may selectively affect tumours with specific molecular characteristics.

“PKMYT1 is attractive because genetic and functional evidence links its inhibition to synthetic lethal vulnerabilities in specific tumour contexts, particularly those involving CCNE1 amplification and alterations in genes such as FBXW7 and PPP2R1A,” Zhavoronkov explains. These genetic alterations affect cell cycle regulation and tumour growth, making affected cancers more dependent on PKMYT1 activity.

PKMYT1 is attractive because genetic and functional evidence links its inhibition to synthetic lethal vulnerabilities in specific tumour contexts, particularly those involving CCNE1 amplification and alterations in genes such as FBXW7 and PPP2R1A

However, PKMYT1 has proved difficult to target with conventional inhibitors.

“At the same time, it’s challenging because prior inhibitor scaffolds have had issues around selectivity, molecular diversity and tolerability. PKMYT1 may also contribute to tumour biology through non-catalytic roles, which can make pure inhibition an incomplete strategy,” he explains.

These characteristics suggest PKMYT1 may benefit from degradation strategies rather than inhibition alone.

Designing a dual-mechanism molecule

The molecule developed in this study combines kinase inhibition with targeted protein degradation using a PROTAC-based design. PROTACs (proteolysis targeting chimeras) are bifunctional molecules that bind a target protein and recruit proteins involved in cellular degradation. The compound both inhibits PKMYT1 activity and directs the protein to the cell’s natural recycling system, leading to its removal.

“This study shows an end-to-end example of using AI to discover a novel PKMYT1 inhibitor chemotype and then extend it into a bifunctional degrader with a dual mechanism, both degrading PKMYT1 and inhibiting its function,” Zhavoronkov says.

The resulting molecule demonstrated promising preclinical activity and may also serve as a research tool for investigating PKMYT1 biology.

The result is a molecule with strong preclinical activity, high selectivity, favourable oral exposure and in vivo antitumour efficacy as a monotherapy. It also provides a useful chemical tool to interrogate PKMYT1 biology in cancer.

“The result is a molecule with strong preclinical activity, high selectivity, favourable oral exposure and in vivo antitumour efficacy as a monotherapy. It also provides a useful chemical tool to interrogate PKMYT1 biology in cancer.”

The dual mechanism may also offer advantages over conventional inhibitors alone.

“A dual mechanism can provide more durable and comprehensive target control,” Zhavoronkov states. “Degradation can drive sustained pathway suppression even after compound washout, while residual inhibitory activity can reinforce functional blockade in conditions where degradation may be incomplete or reduced.”

Together, these properties may improve the consistency of PKMYT1 suppression across different tumour contexts.

Applying generative AI to degrader design

The design process combined generative AI with structure-guided modelling to optimise both the PKMYT1-binding component and the PROTAC linker.

For the inhibitor component, computational design was guided by structural information describing PKMYT1 binding interactions.

“For the warhead, the team used structure-informed pharmacophore concepts and generative chemistry approaches to propose novel PKMYT1-binding molecules, then triaged and optimised candidates through iterative design, modelling and experimental validation to identify a new chemotype,” he explains.

Designing the linker required modelling of the ternary complex formed between PKMYT1, the degrader molecule and the recruited E3 ligase.

“For the linker, a modelled PKMYT1–PROTAC–CRBN ternary complex informed geometry and attachment points,” Zhavoronkov adds. “Generative and structure-guided design then supported the creation and optimisation of linkers that enabled productive degradation while preserving potency and improving developability.”

This approach illustrates how computational design can be applied to the multi-parameter optimisation problems typical of degrader discovery.

Preclinical activity and selectivity

Preclinical studies demonstrated efficient PKMYT1 degradation and downstream pathway modulation in cellular models.

“In cells, the lead PROTAC showed very potent PKMYT1 degradation with high maximal degradation and nanomolar or sub-nanomolar potency, alongside robust downstream pathway effects consistent with PKMYT1 engagement,” Zhavoronkov highlights.

Mechanistic studies confirmed degradation through the ubiquitin–proteasome pathway, the cell’s primary system for breaking down unwanted proteins, with sustained target suppression observed after compound washout. Selectivity was evaluated using kinase profiling and global proteomic analysis, where PKMYT1 emerged as the primary significantly depleted protein under the tested conditions.

Together, these findings support the molecule as both a potential development candidate and a chemical probe for further study.

Challenges in AI-driven PROTAC discovery

Designing PROTAC molecules requires balancing multiple structural and pharmacological constraints.

“PROTAC design is inherently multi-constraint and you need productive ternary-complex formation, strong degradation efficiency, high selectivity and workable properties for exposure and tolerability,” Zhavoronkov notes.

“The team addressed this by combining structure-guided modelling of the ternary complex with systematic E3-binder and linker optimisation, then iterating based on cellular degradation readouts, selectivity profiling and property/PK considerations.”

Future development

The PKMYT1 degrader is currently positioned as both a development lead and a research tool for investigating PKMYT1 biology. Further work will focus on optimisation and biological characterisation.

“Next steps include further optimisation, especially around improving degradation performance across concentrations and contexts, continued selectivity and safety characterisation, along with exploration of rational combination strategies suggested by the biology,” he says.

The research also illustrates a generalisable workflow for degrader discovery.

“More broadly, the work demonstrates a repeatable approach for extending AI-discovered inhibitors into targeted protein degraders, which can be applied to additional oncology targets,” Zhavoronkov concludes.

For early drug discovery, the study demonstrates how generative AI and structure-guided design can be combined to address complex molecular design challenges in targeted protein degradation.