Developing Sustainable Classification of Diseases via Deep Learning and Semi-Supervised Learning.

Area of Science:

• Genetics and Molecular Biology
• Computational Biology
• Machine Learning

Background:

• Disease classification is vital in genetics and molecular biology, typically requiring extensive labeled data.
• Limited labeled samples hinder supervised learning performance in many real-world scenarios.
• Publicly available unlabeled sequence data offers a valuable resource for improving model training.

Purpose of the Study:

• To develop an improved disease classification model using combined deep learning and semi-supervised learning.
• To leverage unlabeled biological data through a self-training mechanism for enhanced model performance.
• To accurately identify disease-causing genes.

Main Methods:

• A novel combined deep learning and semi-supervised learning model was proposed.
• A self-training approach was implemented to generate high-confidence pseudo-labeled samples from unlabeled data.
• The deep forest method was utilized with specific hyperparameter settings.

Main Results:

• The proposed model demonstrated good performance in disease classification tasks.
• The approach successfully improved the utilization of unlabeled samples.
• Effective disease-causing gene identification was achieved.

Conclusions:

• The integration of deep learning with semi-supervised learning and self-training offers a robust solution for disease classification with limited labeled data.
• This method enhances the utility of public sequence databases for biological research.
• The model shows promise for both disease classification and genetic analysis.