More Than Blocking—Sending It for Destruction: AI-Designed Dual-Action Molecule Targets a Cancer Cell-Cycle Vulnerability

Cancer cells’ uncontrolled proliferation may also create dependencies they cannot easily escape. A research team focused on PKMYT1, a kinase that regulates a cell-division checkpoint, and used generative artificial intelligence to design the bifunctional molecule D16-M1P2, seeking both to block the protein’s activity and send the entire protein into the cell’s recycling system. This strategy targets tumors with specific genetic abnormalities rather than broadly attacking all rapidly dividing cells.

PKMYT1 inhibits CDK1, preventing cells from entering division prematurely before DNA replication is complete. Cancer cells with CCNE1 amplification or mutations in FBXW7 or PPP2R1A experience greater replication stress and therefore depend more heavily on this brake. Once PKMYT1 becomes ineffective, the cells may enter division at the wrong time, causing fatal genomic chaos. This vulnerability, which arises only in specific genetic backgrounds, is known as “synthetic lethality.”

Starting with known protein structures and pharmacological characteristics, the researchers used more than 40 AI models on the Chemistry42 platform to generate 2,023 molecules that could potentially bind PKMYT1. Through clustering, redocking, manual review, synthesis, and experimental optimization, they gradually obtained a new binding scaffold. The team then attached it to a structure capable of recruiting the cereblon E3 ubiquitin ligase, creating a proteolysis-targeting chimera (PROTAC): one end of the molecule binds PKMYT1, while the other recruits the cell’s tagging and proteasome systems, prompting clearance of the target protein.

D16-M1P2 is notable for retaining direct inhibitory activity in addition to inducing degradation. Experiments in breast cancer cells showed that it reduced PKMYT1 by as much as approximately 90%; after the drug was washed out, degradation and suppression of downstream CDK1 signaling persisted for at least 24 hours. If the PROTAC’s degradation efficiency declines at high concentrations due to what is commonly called the “hook effect,” the molecule itself can still suppress residual kinase activity, providing a second mechanism of action.

In selectivity testing, D16-M1P2 caused only 4 of 403 kinases, including PKMYT1, to reach the predefined inhibition threshold. Oral administration also produced a dose-related effect in xenograft mice bearing breast cancer with CCNE1 amplification: after twice-daily dosing for 21 consecutive days, the two tested doses inhibited tumor growth by 35.1% and 66.4%, respectively. The study also obtained acceptable pharmacokinetic results in multiple animal species, but these data were primarily intended for early candidate screening and cannot be used to directly infer efficacy or safety in humans.

The findings also draw clearer boundaries around “AI-designed drugs”: algorithms helped explore new scaffolds and linkers, but the actual candidate molecule still emerged from multiple rounds of chemical modification, cellular testing, and animal experiments. More importantly, D16-M1P2’s overall antitumor effect under the study conditions did not clearly surpass that of existing PKMYT1 inhibitors. It currently remains a chemical probe and a pre-candidate lead and has not yet entered human clinical trials. Whether it can be developed into a drug will still require answers regarding long-term toxicity, human exposure, tumor drug resistance, and biomarker-based patient selection.