HEC-ASD: a hybrid ensemble-based classification model for predicting autism spectrum disorder disease genes

Area of Science:

• Genetics
• Computational Biology
• Bioinformatics

Background:

• Autism spectrum disorder (ASD) is a prevalent neurodevelopmental condition with complex genetic underpinnings.
• Current research disproportionately focuses on environmental factors, neglecting the significant 80% genetic contribution to ASD.
• Identifying specific genes responsible for ASD is challenging due to the disease's complexity.

Purpose of the Study:

• To develop a novel hybrid ensemble-based classification (HEC-ASD) model for predicting genes associated with autism spectrum disorder (ASD).
• To leverage gene ontology (GO) and a hybrid gene similarity (HGS) method for improved gene functional analysis.
• To enhance the prediction accuracy of ASD-related genes by employing ensemble gradient boosting machines.

Main Methods:

• The HEC-ASD model was developed using gradient boosting machines for classification.
• A hybrid gene similarity (HGS) method was employed to construct a gene functional similarity matrix utilizing gene ontology (GO).
• The HGS method combines graph-based approaches with GO term hierarchy to effectively measure semantic similarity between genes.

Main Results:

• The HEC-ASD model demonstrated improved classification performance in predicting ASD genes.
• Evaluation on the Simons Foundation Autism Research Initiative (SFARI) gene database showed promising results.
• The model achieved a prediction accuracy of 0.88%, outperforming approaches using gene regulatory networks (GRN) or protein-to-protein interaction networks (PPI).

Conclusions:

• Ensemble learning techniques, specifically gradient boosting, are effective for predicting autism spectrum disorder genes.
• The HEC-ASD model's utilization of gene ontology (GO) offers an effective alternative to protein-to-protein interaction (PPI) networks and gene regulatory networks (GRN) for ASD gene prediction.