Random forest and explainable AI for autism gene prioritization: a machine learning approach to developmental brain data

Objectives Autism Spectrum Disorder (ASD) is a neurodevelopmental condition defined by genetic and environmental factors with causes not fully understood.Methods The study presents a framework utilising Random Forest (RF) models, enhanced by Explainable Artificial Intelligence (XAI) techniques, to rank ASD risk genes using gene expression profiles from the BrainSpan dataset. Through data balancing, including RandomOverSampler, and feature selection such as LASSO, random feature selection, and best-first search, the proposed RF model achieved an average accuracy of 90.9% and an AU-ROC (Area Under the Receiver Operating Characteristic) score of 97.2%, validated using repeated 10-fold cross-validation. We used XAI techniques such as Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) to interpret the model better. LIME offered localised insights by estimating the model output for specific predictions, aiding in identifying crucial gene expression characteristics that influenced classification. Conversely, SHAP provided global interpretability by measuring the marginal contributions of each characteristic across all predictions.Results These techniques elucidated essential gene expression patterns, improving the clarity and dependability of the decision-making process. The proposed RF model has shown enhanced efficacy compared to conventional Support Vector Machines (SVM), especially in detecting protein-coding and non-coding genes potentially linked to ASD.Conclusion This computational paradigm effectively mitigates class imbalance, enhances predictive accuracy, and offers interpretable insights into the genetic underpinnings of ASD. The findings improve ASD diagnosis and provide a basis for gene prioritisation by integrating interpretable machine learning with transcriptome data, yielding translational insights for ASD research and biomedical AI.