top of page

AI Tool for Cancer Therapy Matching Shows Potential, Needs More Data

TAIPEI, TAIWAN, May 31, 2024- A team from Sanford Burnham Prebys in San Diego has developed an AI tool called PERCEPTION, showing promise in predicting how cancer patients will respond to treatments by analyzing single-cell data. However, the tool's progress is slowed by insufficient training data.

Sanju Sinha, the lead developer of PERCEPTION (Personalized Single-Cell Expression-Based Planning for Treatments In Oncology), emphasizes the need for more tumor RNA sequencing data. He urges cancer researchers to share their data openly to refine AI tools for precision medicine.


Current methods for matching patients to drugs often rely on bulk RNA sequencing, which averages gene expression across all cells in a tumor sample. However, tumors consist of various cell subpopulations, known as clones, which can respond differently to treatments. Single-cell RNA sequencing offers higher resolution data, identifying gene expression in individual cells. This method can pinpoint drug-resistant clones that bulk sequencing might miss, but it is expensive and not yet widely available in clinical settings.


AI methods are crucial for managing the massive amounts of data generated by single-cell sequencing. For instance, analyzing 10,000 cells from a single patient can produce around 10 million data points, a volume traditional computational methods struggle with. AI tools like PERCEPTION aim to manage and translate this data for clinical use, but the lack of available training data limits their progress. Sinha advocates for clinicians and researchers to make their single-cell RNA-seq data public, as large datasets are essential for advancing AI tools beyond the proof-of-concept stage.


In a recent study in Nature Cancer, Sinha and his team demonstrated that PERCEPTION could predict therapy responses in multiple myeloma and breast cancer patients. It also identified lung cancer patients resistant to standard treatments, outperforming other predictors based on bulk transcriptomics.


PERCEPTION uses transfer learning, a machine learning technique where a model trained on one task is adapted for a new, similar task. Initially, the model was trained on bulk RNA-seq data to predict drug responses in cell lines. It was then fine-tuned using single-cell RNA-seq and drug response data from the same cell lines. The tool was tested on clinical data from a multiple myeloma study conducted at Tel Aviv Sourasky Medical Center, which included single-cell expression data and treatment responses for 28 patients. The results were promising, showing that PERCEPTION could predict treatment responses effectively.


An important finding from Sinha's study is that the most resistant clones within a tumor determine a patient's overall response to treatment. Even if some cell populations respond well, the presence of highly resistant clones can lead to treatment failure. Conversely, if there are no highly resistant clones, a patient may still achieve a moderate response.


While PERCEPTION shows potential, it is not yet ready for prospective clinical testing. Sinha plans to conduct a larger-scale retrospective validation study. His team is collaborating with researchers locally and internationally to gather more data. Sinha hopes to test PERCEPTION's predictive capabilities in prospective clinical trials if the retrospective validation is successful. Another goal is to use single-cell transcriptomic data to anticipate and counteract the emergence of resistance to therapy.


Sinha and his team have made PERCEPTION available on GitHub, providing instructions for building predictive models in line with promoting data sharing. This openness is crucial for advancing AI in biomedicine and ultimately improving cancer treatment outcomes.


The work on PERCEPTION was conducted by the Cancer Molecular Therapeutics Program at Sanford Burnham Prebys and is supported by collaborators from the National Cancer Institute. By making data more accessible and refining AI tools, researchers hope to enhance the precision and effectiveness of cancer treatments.



Reference:




About WASAI Technology Inc.

WASAI Technology's mission is to deliver acceleration technologies of High-Performance Data Analysis (HPDA) in future data centers for targeted vertical applications with massive volumes and high velocities of scientific data. To strengthen and advance scientific discovery and technological research via big data-intensive acceleration in high-performance computing, WASAI Technology aims to improve commercialization and commoditization of scientific and technological applications.

​​###

Comments


WASAI Tecnology Inc.

4F, No. 6, Zhiyuan 3rd Rd., Beitou Dist., Taipei 112025, Taiwan 

wasai@wasaitech.com

bottom of page