DeepASDPred: a CNN-LSTM-based deep learning method for Autism spectrum disorders risk RNA identification

Area of Science:

• Genetics and Bioinformatics
• Neuroscience
• Computational Biology

Background:

• Autism spectrum disorders (ASD) are neurodevelopmental conditions influenced by genetics, affecting communication and behavior.
• De novo mutations in protein-coding genes are linked to ASD, necessitating efficient identification of risk factors.
• Next-generation sequencing aids in identifying ASD risk RNAs, but current methods are resource-intensive.

Purpose of the Study:

• To develop an efficient computational model for predicting autism spectrum disorder (ASD) risk RNA.
• To leverage deep learning for enhanced accuracy in ASD risk gene identification.

Main Methods:

• Utilized K-mer for RNA transcript sequence feature encoding.
• Integrated gene expression values to create a feature matrix.
• Employed chi-square test and logistic regression for feature selection.
• Developed a binary classification model using convolutional neural networks (CNN) and long short-term memory (LSTM).

Main Results:

• DeepASDPred demonstrated superior performance compared to state-of-the-art methods in tenfold cross-validation.
• The model effectively identifies ASD risk RNA genes.
• The developed predictor achieved outstanding performance in experimental evaluations.

Conclusions:

• DeepASDPred offers a highly effective computational approach for identifying ASD risk RNA.
• The study highlights the potential of deep learning in advancing ASD research.