Low Dimensional Representation of Multi-Patient Flow Cytometry Datasets Using Optimal Transport for Measurable...

Area of Science:

• Hematology
• Computational Biology
• Statistical Learning

Background:

• Accurate quantification of Measurable Residual Disease (MRD) is critical for Acute Myeloid Leukemia (AML) patient prognosis and follow-up.
• Traditional methods lack sensitivity for low leukemia cell detection; flow cytometry datasets offer improved reliability.
• High-dimensional flow cytometry measurements (FCM) from multiple patients present challenges for statistical analysis.

Purpose of the Study:

• To explore statistical learning methods based on optimal transport (OT) for low-dimensional representation of multi-patient FCM datasets.
• To enable effective visualization and clustering of FCM data for MRD assessment in AML.
• To demonstrate the advantages of OT-based methods over existing techniques for analyzing complex FCM data.

Main Methods:

• Utilized optimal transport (OT) framework for dimensionality reduction of large-scale FCM point clouds via K-means mean measure quantization.
• Applied Wasserstein Principal Component Analysis (PCA) and log-ratio PCA for embedding low-dimensional quantized probability measures.
• Validated the approach using public and hospital-based FCM datasets, comparing against kernel mean embedding.

Main Results:

• Demonstrated the superiority of the OT-based approach over kernel mean embedding for statistical learning from multi-patient FCM data.
• Showcased the utility of the methodology for low-dimensional projection and clustering of patients based on MRD levels.
• Achieved relevant and informative two-dimensional representations of FlowSom algorithm results for MRD detection in AML.

Conclusions:

• Optimal transport provides a robust framework for analyzing and visualizing high-dimensional FCM data in AML.
• The proposed OT-based dimensionality reduction and visualization methods enhance the understanding of intra- and inter-patient variability.
• This approach offers significant benefits for MRD quantification and patient stratification in AML management.