Tumor gene expression data classification via sample expansion-based deep learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Tumor diagnosis requires effective methods for discriminating between tumor and normal samples.
• Early tumor detection is crucial for successful patient treatment.
• Traditional methods like Support Vector Machines (SVMs) are used, but deep learning application is limited by insufficient gene expression data.

Purpose of the Study:

• To address the challenge of insufficient training samples in deep learning for tumor gene expression data classification.
• To develop novel deep learning models for accurate tumor classification.
• To enhance the performance of deep learning models by expanding the available gene expression datasets.

Main Methods:

• A Sample Expansion method, inspired by Denoising Autoencoders (DAE), was developed to generate more training samples.
• The applicability of Stacked Autoencoder (SAE) and 1-dimensional Convolutional Neural Network (1DCNN) on gene expression data was analyzed.
• Two deep learning models, Sample Expansion-Based SAE (SESAE) and Sample Expansion-Based 1DCNN (SE1DCNN), were designed and implemented.

Main Results:

• The Sample Expansion method effectively increases the number of available training samples for gene expression data.
• Both SESAE and SE1DCNN models demonstrated high effectiveness in classifying tumor gene expression data.
• The proposed methods successfully overcome the limitation of insufficient training samples.

Conclusions:

• The Sample Expansion method is a viable approach to augment gene expression datasets for deep learning.
• SESAE and SE1DCNN are effective deep learning models for tumor classification using gene expression data.
• This work contributes to advancing computational methods for tumor diagnosis and therapy.