Protein Function Prediction Based on Group Correlation

Some current methods for protein function prediction mainly rely on sequence similarity; however, such methods exhibit limited performance when dealing with sparse data. To address this problem, we propose a novel method for analogous multi-label classification using group correlation. We first construct a PSSM-AC-based feature matrix X. Each row of X corresponds to a PSSM-AC feature of a protein, forming a fixed-length representation. Subsequently, we construct the loss function analogously to the multi-label classification problem and introduce the concept of group correlation into it, shifting the focus of the model from global to local feature correlation to enhance the handling of sparse data. Finally, we conduct comparative experiments to evaluate the proposed method against several commonly used protein function prediction methods. The experimental results demonstrate that our method exhibits significant advantages in handling sparse data. In addition, our method captures the local correlation of intragroup GO terms more efficiently, thereby enhancing the model's performance in protein function prediction. These results validate the effectiveness of our method and provide a new perspective for protein function prediction.