Deep Morphology Learning Enhances Ex Vivo Drug Profiling-Based Precision Medicine

Area of Science:

• Oncology
• Computational Biology
• Genomics

Background:

• Drug testing in patient biopsy-derived cells aids in identifying treatments for relapsed or refractory hematologic cancers.
• Current methods rely on diagnostic markers for cell identification, which can be complemented by morphology-based approaches.

Purpose of the Study:

• To investigate the use of weakly supervised deep learning on cell morphologies (DML) to enhance ex vivo drug screening in hematologic cancers.
• To assess DML's ability to identify malignant and nonmalignant cells and improve drug response prediction.

Main Methods:

• Utilized deep learning on cell morphologies (DML) across 390 biopsies from 289 patients with diverse blood cancers.
• Adapted DML to account for batch effects and autonomously recognize disease-associated cell morphologies.
• Compared DML-based drug recommendations with marker-based and physician's choice treatments in a post hoc analysis of 66 patients.

Main Results:

• DML-based drug responses demonstrated improved reproducibility and clustering of drugs with similar modes of action.
• DML successfully adapted to batch effects and identified disease-associated cell morphologies.
• DML-recommended treatments led to improved progression-free survival compared to other methods.
• Treatments recommended by both immunofluorescence and DML doubled the fraction of patients achieving exceptional clinical responses.

Conclusions:

• DML-based ex vivo drug screening is a promising tool for identifying effective personalized treatments for hematologic cancers.
• Integrating DML into drug screening workflows enhances accuracy and robustness, offering a viable option for clinical routine.
• DML improves treatment selection, potentially leading to better clinical responses and survival rates for patients with advanced blood cancers.