Optimal Supervised Reduction of High Dimensional Transcription Data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-dimensional transcription datasets pose challenges in biological research.
• Complex disorders like cancer involve multigenic traits, amplifying data complexity.
• Dimensionality reduction often involves a trade-off with data integrity.

Purpose of the Study:

• To introduce a novel supervised technique, Class Separation Transformation (CST).
• To address the challenge of dimensionality reduction in high-dimensional transcriptome data for complex traits.
• To enable simultaneous dimensionality reduction and data integrity preservation.

Main Methods:

• Class Separation Transformation (CST) is a supervised technique.
• CST reduces high-dimensional input space to a one-dimensional transformed space.
• The method computes feature importance for class distinction, enabling explainable ML.

Main Results:

• CST achieves optimal class separation in a reduced dimensional space.
• Demonstrated superior accuracy, robustness, scalability, and computational advantage over existing methods.
• Validated using both real and synthetic datasets.

Conclusions:

• CST is an effective method for analyzing high-dimensional transcriptome data.
• The technique offers deeper insights and discovery potential for complex multigenic traits.
• CST provides a robust and computationally efficient solution for bioinformatics challenges.