A new LSTM-based gene expression prediction model: L-GEPM.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profiles are crucial for various biological applications, including disease research and drug discovery.
• Current methods often rely on linear models, neglecting complex nonlinear relationships in gene expression data.
• The high cost and sample requirements for generating comprehensive gene expression datasets remain a challenge.

Purpose of the Study:

• To develop an advanced gene expression prediction model that accounts for nonlinear features.
• To improve the accuracy and fitting effects of gene expression prediction compared to existing methods.
• To leverage deep learning, specifically LSTM networks, for enhanced biological data analysis.

Main Methods:

• Proposed L-GEPM, a gene expression prediction model based on Long Short-Term Memory (LSTM) neural networks.
• Utilized deep learning to capture nonlinear features influencing gene expression relationships.
• Compared L-GEPM against existing linear regression models, analyzing experimental errors and fitting effects.

Main Results:

• The L-GEPM model demonstrated significantly lower prediction errors.
• The LSTM-based approach showed a superior fitting effect for gene expression data.
• Successfully captured nonlinear features, enhancing the predictive power of the model.

Conclusions:

• L-GEPM offers a more accurate and effective approach for gene expression prediction.
• Deep learning, particularly LSTM networks, is well-suited for modeling complex gene expression dynamics.
• This advancement has implications for gene function prediction, crop breeding, and disease-related gene discovery.