← Back to Home

How Can Targeting Cancer Stem Cells Spare Normal Tissue? New Review Proposes an AI-Guided Target Selection Approach

Cancer stem cell markers may change during treatment and may also appear on healthy cells. A new review places AI-designed CAR recognition components within this biological challenge: even if a receptor can be built, it must still be shown to safely identify the cells that should actually be eliminated.

By SURL BioNews

One of the most difficult problems in cancer treatment is that even after a tumor shrinks, cells capable of driving a recurrence may remain. Directing immune cells to pursue these cancer cells with stem cell properties requires sufficiently precise recognition: some candidate markers are also present on normal stem or progenitor cells, meaning the ability to attack tumors could also become a source of damage to healthy tissue.

A review published on November 4 in Clinical Cancer Bulletin, written by Visula Abeysuriya, Sachintha N. Alwis and Vasitha Abeysuriya, proposes a research approach combining single-cell analysis, multiomics data, target screening and computer-assisted design of CAR binding components. A report featured by Bioengineer.org, with Scienmag identified as its news source, introduced the concept. Its focus is on addressing both treatment resistance and safety, rather than announcing a new therapy with proven efficacy.

Cancer stem cells also cannot simply be viewed as a group of cells bearing fixed labels. Their states and surface antigens may change in response to their environment and treatment pressure, allowing previously recognizable cells to evade attack. The review therefore argues that target selection should consider more than which markers are more abundant in tumors. It should also compare expression across different cell states and normal tissues to assess whether there is sufficient room to eliminate cancer cells while protecting healthy cells.

AI has two distinct tasks in this approach: first, integrating single-cell and multiomics data to help identify antigens with greater tumor selectivity; and second, designing the antigen-binding portion of a CAR so that engineered immune cells can recognize the target. The former answers “whom to attack,” while the latter addresses “how to recognize them.” Even successful design of a recognition component does not directly establish that the target itself is sufficiently safe.

The authors specifically caution that existing research on AI-designed CARs has not demonstrated validated cancer stem cell-specific recognition targeting CD133, CD44, EpCAM or CD123. The overlap of these candidate markers with normal cells is precisely the hurdle that must be overcome. Existing achievements in receptor engineering provide a technical basis, but remain insufficient to demonstrate that cells capable of regenerating tumors can be safely eliminated in patients.

The report also discusses the possibility of using AI to continuously track changes during treatment and creating CAR-T “digital twins”: updating patient data to simulate interactions between tumors and immune cells and explore how treatment resistance develops. However, these remain developing concepts and have not yet become mature clinical tools that can guide adjustments to cancer stem cell treatment.

This review outlines a research framework that remains to be validated. Most applications are currently still at the computational or preclinical stage, and prospective clinical benefit has not yet been established. To bring AI-selected targets into treatment, researchers must still use experimental and clinical data to answer the same question: can they keep pace with changes in cancer cells while maintaining a sufficiently reliable margin of safety?

References

  1. Bioengineer.org
  2. Clinical Cancer Bulletin
  3. Scienmag