Artificial intelligence and precise “reprogramming” of tumor cells bring new hope in the treatment of colorectal cancer: in a series of experiments, researchers managed to force the most resistant part of the tumor – cancer cells with stem cell properties (CSCs) – to lose their malignant identity and then self-destruct. At the heart of the approach is the CANDiT algorithm which, starting from a single key point in the tumor's genome, proposes “switches” to return the cells to a more differentiated state. When this happens, the cells that sustained the tumor and fueled its recurrence can no longer survive. The research is focused on colorectal cancer, but the methodology is designed to be tumor-agnostic and transferable to other sites.

Why CSCs are the hard core of the tumor

CSCs are not numerous, but they are crucial to the nature of the disease. They drive metastases, evade chemotherapy and immunotherapy, hide through minimal residual disease, and re-initiate growth after apparent cure. In colorectal cancer, CDX2, a transcription factor that maintains the differentiation program in a healthy intestine, holds a special place in this story. When CDX2 is low or absent, the tumor is more aggressive, therapeutic choices are more complex, and the risk of recurrence is higher. By introducing precise algorithmic tools, it is possible to map where this program "broke" and how to restore it.

CANDiT: the algorithm that finds the switch for differentiation

CANDiT (Cancer-Associated Nodes for Differentiation Targeting) starts from a single reference point – a gene that is essential for healthy tissue growth but is silenced in aggressive tumors. For colorectal cancer, researchers chose CDX2. Then, in vast datasets encompassing thousands of human tumors, they built a network of genes and pathways that are functionally linked to CDX2. The idea is simple but powerful: not just looking for a single mutated gene, but looking for a network node whose activation can trigger a return to normal, i.e., differentiation.

The model singled out an unexpected target: PRKAB1, the beta-subunit of the AMPK complex, a central sensor of cellular energy and stress. When scientists activated PRKAB1 with an available agonist, they achieved CDX2 re-expression and activation of signatures accompanying intestinal differentiation. The result was selective: the effect was clearly seen in CDX2-low tumors, while healthy tissue remained largely spared. It is precisely this selectivity that carries therapeutic value – targeting cells dependent on the cancerous program, not the surrounding structures.

From silicon to “mini-tumors”: proof on patient-derived organoids

CANDiT's predictions did not remain on computer screens. At the HUMANOID™ Center at UC San Diego, the team tested the hypothesis on patient-derived tumor organoids – small, three-dimensional replicas that preserve the architecture, heterogeneity, and behavior of the real tumor. Organoids are a key link because they allow for safe and rapid verification of whether a drug works in human tissue: measuring the transcriptome, metabolism, proliferation, signs of differentiation, and, finally, selective cytotoxicity in the very cells that sustain the disease.

It was these experiments that showed the most interesting phenomenon: when CDX2 is restored through reprogramming and the differentiation program is "tightened," the CSCs not only lose their stemness but also initiate programmed cell death. It is as if, deprived of their cancerous identity and the energetic "switch" they used to survive, they can no longer persist. This opens a new therapeutic horizon: instead of just blocking CSCs, we can induce them to turn themselves off.

What is the potential benefit and for whom

To assess the potential real-world benefit, the authors built a gene signature of response – a recognizable pattern of activation and silencing of multiple genes – that predicts who will best respond to therapy aimed at restoring CDX2. Then, in computational simulations mirroring the diversity of large clinical trials, they tested the signature in a dozen independent cohorts with more than 2,000 patients. The analyses suggest that in selected colorectal cancers, with the activation of PRKAB1 and the return of CDX2, the risk of recurrence and death could be reduced by up to 50%. It is important to emphasize: this is a biomarker-driven therapy that targets a specific biological vulnerability, not all patients equally.

The advantage of repurposing: a candidate drug we already know

A major practical advantage of the approach is that it is not necessary to invent a completely new drug. AMPK activators, including compounds targeting PRKAB1, already have some safety and pharmacodynamic data from earlier development phases. This shortens the path to the clinic and increases the likelihood that initial human trials can begin with clear biomarker endpoints: CDX2 restoration, transcriptome changes, and a decrease in tumor burden. At the same time, such drugs offer combinatorial potential with existing therapies – cytostatics, anti-EGFR antibodies, and immunotherapies – especially in resistance scenarios.

The broader context of 2025: where the new strategy fits in

The year 2025 brings several important lessons in colorectal cancer: progress in subgroups with the BRAF V600E mutation, more precise combinations of targeted drugs, and the increasing role of organoids and multidimensional "omics" in predicting response. Algorithm-guided differentiation therapy introduces a third therapeutic axis – targeting the cellular state – which complements mutational targets and immune niches. In practice, this could mean that after standard surgery and systemic therapy, biomarker-positive patients are enrolled in adjuvant or conversion protocols that disarm the CSC population.

What the path to the clinic looks like: steps and benchmarks

Before the concept becomes routine, a series of necessary steps must follow. First, GLP toxicology and pharmacokinetics of optimized PRKAB1 agonist variants to achieve stable and predictable exposure. Second, first-in-human trials in a carefully selected population (CDX2-low tumors with a positive response signature), with embedded biomarker-targets and early readouts of effect (e.g., changes in circulating tumor DNA, CDX2 restoration in biopsy, patient organoid dynamics). Third, randomized trials to show whether the modeled benefit (reduction in recurrence and death risk by up to 50%) can be reproduced in a real-world clinical population.

HUMANOID™: “Phase Zero” – a clinical trial in the lab

The foundation for faster progress is an infrastructure that allows for the simulation of a clinical trial in laboratory conditions. At the HUMANOID™ Center, patient-derived organoids serve as phase zero: they measure not only "does the drug work," but also how precisely it works, with safety frameworks in place before it reaches the patient. Such a cycle – from algorithm, to organoid, to validation in larger datasets – shortens development time from years to months and reduces the risk of expensive late-stage failures.

To whom and when: patient selection and operational implementation

Introducing this strategy into practice will require clear inclusion criteria. In the clinic, it would look like this: after standard histopathological workup, CDX2 testing and assessment of the gene signature of response are performed. Patients whose tumors show CDX2-low status and a positive signature of probable response would be candidates for inclusion in a trial of a PRKAB1 activator. During treatment, early biomarkers would be monitored (transcriptome changes, drop in ctDNA, differentiation markers in control biopsies), and the algorithm would iteratively guide dosing and combinations.

Safety and selectivity: what distinguishes this approach from cytotoxic chemotherapy

Unlike standard cytostatics, which act on everything that divides rapidly, CANDiT-guided differentiation therapy targets the tumor's functional dependence on specific network nodes. In experiments, the effect was target-specific: when PRKAB1 is genetically silenced, the pharmacological agonist loses its effect, confirming that the therapy indeed hits the predicted target. This also explains the relatively favorable safety profile compared to non-specific cytotoxic approaches, and it opens up space for smart combinations that enhance differentiation without additional damage to healthy tissue.

Open questions and research frontiers

• Mechanism of self-destruction: The exact sequence of events leading to the death of reprogrammed CSCs needs to be dissected – the role of energy stress, epigenetic "locking" of differentiation, and signaling axes that activate apoptosis.


• Stability of the effect: How long the new, differentiated state lasts and how to prevent a return to "fetal" and stemness programs under therapeutic pressure.


• Microenvironment and immunology: Organoids do not contain all elements of the tumor ecosystem; therefore, co-cultures with immune cells, stroma, and vascular components, as well as in vivo confirmations, will be crucial.


• Tumor heterogeneity: Network signatures can vary within a single tumor; a combination of multiple biopsies and spatial transcriptomics will reduce the risk of misclassification.


• Combination therapies: Synergies with anti-EGFR antibodies, immunotherapies, and drugs targeting mitochondrial function should be tested, especially in subgroups defined by proteomic and spatial profiles.

What patients can do today, October 22, 2025

For those affected by colorectal cancer, the most important thing is to talk to their oncologist about CDX2 testing and other biomarkers, monitor opportunities to enroll in studies targeting tumor plasticity, and, where available, regularly monitor circulating tumor DNA as an early indicator of response. Modern care also includes an organized approach to nutrition, exercise, and post-therapy rehabilitation, which has shown benefits for outcomes and quality of life. Meanwhile, the development of differentiation strategies continues within the framework of strictly controlled protocols that combine algorithmic insights, organoid tests, and clinical verification.

Documents and additional materials

More about the “Phase 0” approach and the infrastructure for rapid therapy validation is available at this link. You can open the preprinted, publicly available version of the scientific paper with methods and results here. For a broader context on the role of CDX2 and modern response signatures in colorectal cancer, we recommend reviewing specialist publications and recent reports from professional congresses.