Deep Few-Shot Network for Protein Family Classification: Bridging the Gap Between Limited Data and High Performance

Protein sequence analysis presents a significant challenge in bioinformatics, impacting applications such as disease investigation and precision medicine. Despite advancements in sequencing technologies and the resultant proliferation of databases, classifying protein families remains a hurdle. This study introduces a novel deep few-shot network specifically designed for protein family classification, addressing the limitations of traditional methods by utilizing deep learning with sparse training datasets. Our experiments show that this framework outperforms state-of-the-art baseline models, achieving 97.3 % precision and 95.2 % recall on the training set, and 95.0 % precision and 93.5 % recall on the test set, when trained with only 25 shots per class. This work represents the first endeavor in crafting a deep network tailored for primary sequence family classification, achieving remarkable performance with minimal observations.