Microfluidic Machine Learning–Driven Platform Predicts Drug Sensitivity in Pediatric T-ALL

Researchers from the University of Utah, St. Jude Children’s Research Hospital, and the University of Pennsylvania have developed the “lab-on-a-chip” μPharma pharmacotyping platform, which combines microfluidics and AI to rapidly predict responses to targeted therapies at the single-cell level in pediatric T-cell acute lymphoblastic leukemia (T-ALL). Findings published in the journal Med suggested that, while not yet used clinically, the automated microdevice could enable same-day precision oncology decision-making.

Current pharmacotyping methods require prolonged drug incubations, extensive manual handling, and have analytical limitations (eg, overlooking single-cell characteristics), which the investigators wrote hamper their clinical feasibility. These methodologic constraints are relevant for pediatric T-ALL, an aggressive disease with limited therapeutic options and lacking actionable genomic markers.

In the present analysis, the AI-driven tool appeared to accurately predict sensitivity to dasatinib and venetoclax—targeted therapies under investigation for T‑ALL—within hours and showed that single-cell spatial protein distribution and morphology enhance forecasting precision.

“Our team has worked hard to develop this technology, and seeing it perform well is a key step toward bringing it into the clinic to help patients,” commented Yue Lu, PhD, of the University of Utah, Salt Lake City, and the Hunstman Cancer Institute, one of the corresponding authors, in an institutional press release.

From a clinical standpoint, Luke Maese, DO, also of both the University of Utah and the Huntsman Cancer Institute, who treats pediatric patients with leukemia who could benefit from μPharma, commented in the press release, “Innovation in treatment selection is a pressing need within pediatric malignancies. Personalized treatment selection accomplished in ‘real-time’ will be part of the future of cancer therapeutics, and μPharma represents an encouraging step in that direction.”

Model Methods

μPharma is a pharmacotyping platform that predicts single-cell drug sensitivity without directly exposing cells by measuring pretreatment biomarkers linked to therapeutic response. It does so using an automated digital microfluidic immunofluorescence assay for suspension cells paired with machine learning models trained on comprehensive single-cell features.

In operation, patient cancer cells are placed into the device, where they are held between two plates. Digital microfluidics move droplets of reagents across the chip to and from the cells, automating liquid-handling tasks, reducing cell and reagent requirements, minimizing human error, and reducing processing time from several days or weeks to approximately 4 hours.

μPharma was validated using T-ALL cell lines and patient-derived xenografts, predicting sensitivity to dasatinib and venetoclax by quantifying their respective target proteins, LCK and BCL2, along with protein expression, phosphorylation status, spatial distribution, and cellular morphology.

Key Findings

The study confirmed that phosphorylated LCK predicts dasatinib sensitivity and identified phosphorylated BCL2 as a previously unrecognized biomarker for venetoclax sensitivity.

Integration of multiple biomarkers into machine learning models, rather than single-marker analyses, appeared to significantly enhance predictive accuracy. The researchers highlighted spatial protein distribution and integrated protein–morphology metrics as key informative features. Additionally, single-cell analysis revealed distinct cell subpopulations, which the researchers noted suggest intratumor heterogeneity in drug responses.

The researchers concluded, “μPharma provides rapid (4-[hour] assay), accurate, and automated prediction of drug sensitivity at single-cell resolution using minimal clinical samples….”

Highlighting the role of μPharma in advancing safer, more precise care, Makala Pace, PharmD, BCOP, MBA, Pharmacy Director at Huntsman Cancer Institute, said in the press release, “A tool that can predict drug response in hours and help clinical teams prioritize therapies with the best chance of benefit—while streamlining care and minimizing unnecessary toxicity for our youngest [patients with cancer]—is exactly the kind of precision we strive for in oncology pharmacy practice.”

In the press release, corresponding author, Alphonsus H. C. Ng, PhD, also of the University of Utah, added, “If we can rapidly and accurately monitor the sensitivity of cancer cells and tailor treatment appropriately, we believe it can significantly improve outcomes.” He continued, “The next step is validation of this technology using primary leukemia cells in a realistic clinical environment.”

DISCLOSURES: The study was funded by institutional start-up funds from the University of Utah, including internal supplements provided through the Immunology, Inflammation, and Infectious Disease Initiative and the Diabetes and Metabolism Research Center. For full disclosures of the study authors, visit cell.com.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.