AI Tool Predicts Personalized Cancer Drug Responses Using Single-Cell Data

Artificial intelligence models can now predict patient responses to breast-cancer drugs by analyzing single-cell RNA sequencing data from tumor clones. Developed by researchers at the National Cancer Institute, the computational pipeline known as PERCEPTION matches specific targeted therapies to individual cellular subpopulations, offering a leap forward in precision oncology.

In Plain English: The Clinical Takeaway

• Single-Cell Resolution: Traditional bulk sequencing averages all cells in a tumor. This new AI tool looks at distinct subgroups of cells, called clones, which often drive drug resistance.
• Transfer Learning: The algorithm was initially trained on large-scale drug screens using standard bulk gene expression data, then refined with high-resolution single-cell data.
• Clinical Relevance: Published findings in Nature Cancer demonstrate that if even a single clone within a tumor is resistant to a medication, the patient may fail to respond, a nuance the PERCEPTION model aims to catch.

Bridging Single-Cell Biology and Clinical Oncology

For decades, oncology treatment planning relied heavily on bulk tumor profiling. According to data from the National Cancer Institute (NCI), bulk sequencing aggregates the genetic material of an entire tumor sample into a single average readout. However, malignant neoplasms are rarely homogeneous. They consist of diverse cellular subpopulations, or clones, each carrying unique mutations and varying phenotypic behaviors.

This cellular heterogeneity explains why a regimen that shrinks a primary tumor might leave behind a drug-resistant clone, leading to relapse. To solve this clinical blind spot, researchers developed PERCEPTION (Personalized Single-Cell Expression-based Planning for Treatments In Oncology). As detailed in research published in Nature Cancer, the computational pipeline evaluates gene expression profiles at single-cell resolution to forecast therapeutic efficacy.

Methodology and Validation Across Multiple Cancers

The research team, led by investigators at the NCI’s Center for Cancer Research including Alejandro Schaffer, Ph.D., Sanju Sinha, Ph.D., and Eytan Ruppin, M.D., Ph.D., approached the data scarcity problem through machine learning. Single-cell RNA sequencing remains costly and less accessible in routine clinical workflows compared to bulk sequencing. To overcome this, the group utilized transfer learning—a technique where a model trained on one large dataset is fine-tuned on a more specialized, smaller dataset.

Photo: cancer.gov

The team constructed predictive AI models for 44 Food and Drug Administration (FDA)-approved cancer drugs using published cell-line screens. They then tested the pipeline on clinical trial cohorts:

Clinical Validation Datasets for the PERCEPTION Pipeline

Cancer Type	Patient Cohort Size	Treatment Type
Multiple Myeloma	41 patients	Combination of four drugs
Breast Cancer	33 patients	Combination of two drugs
Non-Small Cell Lung Cancer	24 patients	Targeted therapies

In the breast cancer cohort of 33 patients, the algorithm accurately anticipated therapeutic responses by identifying minority clones that exhibited innate resistance to specific therapeutic agents.

Regulatory and Translational Challenges in Modern Healthcare

Contraindications & When to Consult a Doctor

Future Outlook

References

• National Cancer Institute. (2024). AI tool helps predicts patient responses to cancer drugs. NCI Press Releases.
• Sinha, S., et al. (2024). Predicting patient response and resistance to treatment from single-cell transcriptomics of their tumors via the PERCEPTION computational pipeline. Nature Cancer, 5, 680–693. 10.1038/s43018-024-00751-4.

Disclaimer: This article is for informational purposes only and does not constitute formal medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified health provider with any questions regarding a medical condition.