A New Approach Based on Feature Selection of Light Gradient Boosting Machine and Transformer to Predict circRNA-Disease Associations

Circular RNA (circRNA) is a type of single-stranded with closed circular structure. Recent studies have shown that circRNA has relatively more stable structure than its linear counterparts. The become biological marker in medicine and plays crucial role disease prediction. However, traditional experiments are often time-consuming laborious. More researchers taking computational approaches to predict the circRNA-disease associations rapidly reliably. In this paper, we propose novel method for predicting based on feature selection using Light Gradient Boosting Machine (LightGBM) self-attention neural network-Transformer (LGFRCDA). Firstly, histogram-based decision tree algorithm LightGBM used discretize continuous floating-point features into histogram integer numbers. While traversing samples, difference between histograms optimize calculation, greatly improving construction speed. Then leaf-wise employed calculate node maximum split gain, resulting final vector. Finally, these sorted order importance introduced Transformer information fusion Our study demonstrates after processing dimension reduction, LGFRCDA achieved prediction accuracy 95.44% AUC (Area Under receiver operating characteristic Curve), which 3.11% higher latest algorithms same dataset. We also conducted search published literature cross-validate predicted result. Out top 15 pairs by model, 13 were confirmed existing literature. These results indicate proposed model suitable can provide reliable candidates experiments.