Prediction and prioritization of autism-associated long non-coding RNAs using gene expression and sequence features

Area of Science:

• Genetics
• Neuroscience
• Bioinformatics

Background:

• Autism spectrum disorders (ASD) are complex neurodevelopmental conditions with a significant genetic component.
• While protein-coding genes are known ASD risk factors, the role of long non-coding RNAs (lncRNAs) in ASD pathogenesis remains largely unexplored.
• Altered expression of some lncRNAs has been observed in autistic brains, highlighting their potential involvement.

Purpose of the Study:

• To develop a novel machine learning approach for predicting candidate lncRNAs associated with ASD.
• To leverage knowledge from protein-coding ASD risk genes to enhance lncRNA prediction.
• To identify and prioritize lncRNAs for further investigation into their roles in ASD.

Main Methods:

• Utilized machine learning, including autoencoder networks for representation learning of gene expression data.
• Applied random-forest-based feature selection to k-mer features derived from transcript sequences.
• Employed logistic regression, support vector machine, and random forest models for prediction and prioritization.

Main Results:

• Developed robust predictive models for ASD-associated lncRNAs using developmental brain gene expression and transcript sequence data.
• Demonstrated effective knowledge transfer from protein-coding ASD risk genes to lncRNA prediction.
• Generated a prioritized list of candidate lncRNAs, including those potentially regulating known ASD risk genes.

Conclusions:

• ASD risk genes can be accurately predicted using developmental brain gene expression and transcript sequence features.
• The developed models offer valuable insights for the functional characterization of ASD-associated lncRNAs.
• This approach aids in identifying novel genetic factors contributing to ASD pathogenesis.