Human Splice-Site Prediction with Deep Neural Networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate splice-site prediction is crucial for gene structure delineation from sequence data.
• Deep neural networks (DNNs) have demonstrated superior performance in various classification tasks, including biological sequence analysis.

Purpose of the Study:

• To propose a novel method for splice-site prediction utilizing deep neural networks (DNNs).
• To evaluate the performance of the proposed DNN-based method against existing techniques.

Main Methods:

• A deep neural network (DNN) model incorporating convolutional and bidirectional long short-term memory layers was developed.
• The model was pretrained and validated on established datasets used in prior splice-site prediction studies.
• Input sequences of 140 nucleotides, centered with consensus splice sequences (GT/AG), were processed by the DNN.

Main Results:

• The proposed DNN method demonstrated superior performance compared to previously reported methods.
• Visualizations using position frequency matrices (PFMs) revealed patterns learned by the DNNs, some resembling consensus sequences.
• The trained DNN model and source code are made available for further research.

Conclusions:

• The developed DNN-based approach offers an effective and improved method for splice-site prediction.
• Further advancements in DNN architectures hold promise for enhanced gene structure analysis.
• The availability of the model and code facilitates reproducibility and future development in the field.