Identification of human circadian genes based on time course gene expression profiles by using a deep learning method

Area of Science:

• Genomics
• Computational Biology
• Chronobiology

Background:

• Circadian genes regulate ~24-hour biological rhythms crucial for human health.
• Current gene identification algorithms suffer from high false positives and low coverage.
• Accurate identification of circadian genes is essential for understanding biological control.

Purpose of the Study:

• To develop a novel computational framework for accurate circadian gene identification using deep learning.
• To overcome limitations of existing methods in detecting periodic genes.
• To identify novel circadian genes and elucidate their functional roles.

Main Methods:

• Constructed a computational framework using deep neural networks (DNN).
• Transformed time-course gene expression data into categorical-state data.
• Clustered state data to identify distinct expression patterns and trained DNN to discriminate gene types.
• Compared DNN performance against k-nearest neighbors, logistic regression, naïve Bayes, and support vector machines.

Main Results:

• The DNN model demonstrated superior balanced precision and recall compared to other machine learning methods.
• Identified 1132 novel periodic genes by applying the DNN model to large-scale transcription profiles.
• Functional analysis revealed distinct circadian expression patterns in the GTPase superfamily, potentially regulating immune system function.

Conclusions:

• The novel DNN framework significantly enhances circadian gene identification accuracy.
• Discovery of novel periodic genes, including GTPase superfamily members, provides new insights into circadian biology.
• This work advances the field of circadian gene identification and understanding of circadian control in human health.