Data-driven characterization of molecular phenotypes across heterogeneous sample collections.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Large gene expression datasets offer potential for understanding disease mechanisms and patient stratification.
• Integrating multi-study and multi-platform data is crucial for robust analysis but requires advanced tools.

Purpose of the Study:

• To develop an intuitive, data-driven approach for molecular profile stratification.
• To benchmark the methodology using t-distributed stochastic neighbor embedding (t-SNE) on hematological malignancy data.
• To enable comparison across datasets and data types for enhanced biological insight.

Main Methods:

• Utilized t-distributed stochastic neighbor embedding (t-SNE) for dimensionality reduction.
• Applied an integrative approach to multi-omics data from acute myeloid leukemia studies.
• Assessed biological vs. technical variation in sample clustering and incorporated additional datasets.

Main Results:

• Identified novel molecular subtypes in acute myeloid leukemia, including a myelodysplastic syndrome-like cluster.
• Discovered a new cluster characterized by CEBPA mutations and altered DNA methylation pathways.
• Demonstrated differential survival and drug responsiveness in identified subtypes, particularly in samples lacking fusion genes.

Conclusions:

• Integrating data across multiple studies is key to discovering novel molecular disease subtypes.
• The developed approach enhances understanding of disease biology and aids in patient stratification.
• This methodology provides a framework for comparative analysis of molecular profiles across diverse datasets.