On transformative adaptive activation functions in neural networks for gene expression inference.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The National Institutes of Health (NIH) Library of Integrated Network-Based Cellular Signatures (LINCS) program developed cost-effective gene expression profiling using landmark genes.
• The Database for Annotation, Visualization and Integrated Discovery (DAVID) Gene Expression (D-GEX) method uses neural networks to infer whole expression profiles from landmark genes.
• Existing D-GEX methods have limitations and can be significantly improved for greater accuracy.

Purpose of the Study:

• To propose a novel transformative adaptive activation function for enhanced gene expression inference.
• To improve the accuracy and efficiency of reconstructing full gene expression profiles from limited landmark gene data.
• To generalize and improve upon existing adaptive activation functions used in neural network-based gene expression profiling.

Main Methods:

• Implementation of a novel transformative adaptive activation function within a neural network architecture.
• Utilizing the D-GEX framework with a focus on inferring entire gene expression profiles from approximately 1,000 landmark genes.
• Comparison of the proposed method's performance against a reimplementation of the original D-GEX algorithm.

Main Results:

• The improved neural network achieved an average mean absolute error (MAE) of 0.1340.
• This represents a significant improvement compared to the reimplemented original D-GEX, which had an MAE of 0.1637.
• The novel activation function demonstrated more accurate reconstruction of full gene expression profiles with minimal increase in model complexity.

Conclusions:

• The proposed transformative adaptive activation function significantly enhances gene expression inference accuracy.
• This advancement offers a more precise and cost-effective approach to gene expression profiling using landmark gene data.
• The method provides a valuable improvement for systems biology and drug discovery applications.