Explore Research

From Single-Cell Data to Biomarker Panels 

Author: Razie Yousefi

Biomarker discovery often begins with identifying genes that distinguish disease states. However, translating these discoveries into robust clinical biomarker panels requires more than statistical significance. Biomarkers must be biologically meaningful, reproducible across independent datasets, robust to technical and biological variability, and measurable in clinically relevant samples.

I am exploring this problem using multiple myeloma (MM) as a model system. The objective is to develop a computational framework that moves from high-resolution single-cell biology toward compact and robust biomarker panels.

I analyze independent single-cell RNA-sequencing datasets spanning the healthy, MGUS, SMM, and MM continuum. Rather than merging datasets, each cohort is analyzed independently to identify reproducible molecular patterns associated with disease progression. Candidate genes are then evaluated in independent bulk bone-marrow transcriptomic data to determine whether signals discovered at single-cell resolution remain detectable at the tissue level. 

The resulting candidate biomarkers are further evaluated using statistical and machine-learning approaches designed to identify stable and informative combinations of features, rather than simply selecting the genes with the strongest statistical significance. Where appropriate, genes representing related cellular populations or biological programs can be combined into higher-level features to improve robustness. 

The broader goal is to develop biomarker panels that bridge three levels of evidence: 

Single-cell biology → patient-level molecular signatures → clinically relevant biomarker panels

This work is part of our broader research focus on combining computational biology, statistical inference, and machine learning to address the challenges of translating complex biological data into practical and reproducible measurements.