Interpretable, Scalable, and Transferrable Functional Projection of Large-Scale Transcriptome Data Using Constrained Matrix Decomposition | JoVE Visualize

Area of Science:

Computational Biology
Systems Biology
Genomics

Background:

Large-scale transcriptome data, including single-cell RNA sequencing, offers systems-level insights into biological processes.
Existing dimensionality reduction methods often lack interpretability, transferability, and direct gene contribution analysis.
Inferring functional variations from gene sets is crucial but challenging with reduced dimensions.

Purpose of the Study:

To develop and evaluate novel dimensionality reduction methods for transcriptome data analysis.
To create interpretable, transferable, and gene-centric low-dimensional representations of transcriptomic variations.
To assess the performance of gene set non-negative principal component analysis (gsPCA) and non-negative matrix factorization (gsNMF) for biological process analysis.

Main Methods:

Application of gene set non-negative principal component analysis (gsPCA) and non-negative matrix factorization (gsNMF) to large-scale transcriptome datasets.
Comparative analysis against existing functional variation inference methods.
Validation of predictive capabilities on unseen experimental conditions, including epithelial-mesenchymal transition (EMT).

Main Results:

gsPCA and gsNMF provide quantitative, low-dimensional insights into biological process progression, comparable to existing methods.
These methods enable accurate predictions of functional space locations for unexposed data, including EMT progression and reversion.
Conserved EMT programs were identified across diverse cell types and tumor samples.
The approach demonstrated broad applicability beyond EMT, with recommendations for method selection and parameter optimization.

Conclusions:

Constrained matrix decomposition methods (gsPCA, gsNMF) yield functionally interpretable and transferable low-dimensional transcriptome data.
These methods offer a powerful, broadly applicable tool for analyzing large-scale transcriptome datasets.
The developed approach enhances the understanding of biological processes and gene contributions in complex transcriptomic data.