Predicting protein interactions for drug target discovery using stacked denoising autoencoders and random ferns

Yang Li^1*Yan-Qiu Wang²ChangQing Yu¹Chen Jiang¹Tailong Shi¹Sizhe Liang¹

• ¹Xijing University, Xi'an, China
• ²Zaozhuang University, Zaozhuang, China

Protein-protein interactions (PPIs) are crucial for understanding disease mechanisms and identifying drug targets. While experimental methods have been widely used to identify PPIs, they tend to be timeconsuming, labor-intensive, and often result in high false discovery rates. This situation has led to an increasing demand for reliable computational approaches to detect PPIs, particularly for advancing drug target discovery. This paper proposes an efficient computational framework, Stacked Denoising Autoencoder with Random Ferns (SDAERFs), for predicting PPIs from protein sequence information.Our model leverages evolutionary information encoded in Position-Specific Scoring Matrices (PSSMs) and employs a stacked denoising autoencoder (SDAE) to extract high-level features. These features are then used by a Random Ferns (RFs) classifier to predict PPIs. The SDAERFs model was extensively validated on the two benchmark datasets, with accuracies of 98.13% and 98.60%, respectively. The results demonstrate the model's validity and reliability in PPIs prediction. Furthermore, we performed comprehensive comparisons with current methods, confirming the superior performance of SDAERFs. Our findings suggest that the SDAERFs model is a powerful tool for PPIs prediction, providing an efficient and reliable approach for advancing drug target discovery and therapeutic development.