A Deep Neural Network Model using Random Forest to Extract Feature Representation for Gene Expression Data...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Gene expression data presents a "n << p" challenge where samples are fewer than features.
• This imbalance hinders traditional deep learning methods in gene expression classification.
• Feature sparsity and unknown correlations add complexity to classification tasks.

Purpose of the Study:

• To introduce a novel classifier, Forest Deep Neural Network (fDNN), for gene expression data.
• To address the limitations of existing methods in handling high-dimensional, low-sample-size data.
• To improve classification accuracy and feature selection in predictive modeling.

Main Methods:

• Integration of a deep neural network architecture with a supervised forest feature detector.
• Development of a method to learn sparse feature representations.
• Utilizing built-in feature detection to mitigate overfitting in neural networks.

Main Results:

• fDNN demonstrated superior classification performance compared to standard random forests and deep neural networks.
• The method effectively learned sparse feature representations from gene expression data.
• Simulation experiments and real-world RNA-seq dataset analyses validated fDNN's capabilities.

Conclusions:

• fDNN is a valuable addition to predictive modeling for gene expression data.
• The classifier offers improved performance and more meaningful feature selection.
• fDNN effectively tackles the "n << p" problem and feature sparsity in bioinformatics.