DeepDSSR: Deep Learning Structure for Human Donor Splice Sites Recognition.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Alternative splicing in humans generates diverse mRNA and protein isoforms.
• Accurate splice site identification is essential for understanding gene structure and functional variants.
• Existing computational methods for splice site detection require further improvement.

Purpose of the Study:

• To develop a novel deep learning-based computational method for predicting human donor splice sites.
• To enhance the accuracy of splice site recognition in human pre-messenger RNA.
• To provide a robust tool for analyzing gene structure and alternative splicing.

Main Methods:

• Development of DeepDSSR (deep donor splice site recognizer), a deep learning architecture.
• Training and evaluation using a publicly available, imbalanced benchmark dataset.
• Comparative analysis against leading deep learning-based methods for donor splice site detection.

Main Results:

• DeepDSSR demonstrates comparable performance to existing leading deep learning methods.
• Performance evaluation metrics indicate that DeepDSSR outperforms current deep learning-based approaches.
• The model effectively addresses the challenge of predicting human donor splice sites.

Conclusions:

• DeepDSSR represents a significant advancement in the computational prediction of human donor splice sites.
• The developed method offers improved accuracy over existing deep learning techniques.
• Future work will focus on enhancing predictive capabilities and developing acceptor splice site prediction.