t-SNE

t-SNE maps high-dimensional molecular data into a low-dimensional space to visualize and preserve local structure for characterization of molecular phenotypes.

Key Features:

• Dimensionality Reduction: Reduces complex gene expression datasets to two or three dimensions to facilitate visualization and interpretation of sample-level patterns.
• Data Integration: Supports integrative analysis across multiple studies and platforms for comparative analysis of molecular profiles.
• Biological vs. Technical Variation Assessment: Enables evaluation of biological variation versus technical artifacts in sample clustering.
• Incorporation of Additional Datasets: Permits direct incorporation of additional datasets into an existing low-dimensional representation.
• Comparison of Molecular Subtypes: Allows comparison of molecular disease subtypes identified from separate t-SNE representations.
• Pathway Database Integration: Enables characterization of clusters using pathway databases and supplementary data to infer underlying biological mechanisms.

Scientific Applications:

• Elucidating Disease Mechanisms: Integrates large gene expression datasets to uncover disease mechanisms and changes in cellular pathways.
• Patient Stratification: Stratifies patients based on molecular profiles for personalized-medicine–oriented analyses.
• Identification of Novel Subtypes: Reveals novel molecular subtypes with distinct clinical features, including differential survival and drug responsiveness.
• Acute Myeloid Leukemia Subtype Identification: Applied in integrative multi-omics analyses of acute myeloid leukemia to identify a myelodysplastic syndrome-like cluster and a CEBPA-mutated cluster with altered S-adenosylmethionine-dependent DNA methylation pathway activity.

Methodology:

t-SNE was applied for data-driven stratification on benchmarked multi-study, multi-platform hematological malignancy datasets, including assessment of biological versus technical variation, incorporation of additional datasets into existing low-dimensional representations, and cluster characterization using pathway databases and supplementary data.