DPPred-indel:a pathogenic inframe indel prediction method based on biological language models and fusion of DNA and protein features

Inframe insertion/deletion (indel) mutations play a crucial role in human genetic variation and are closely associated with various diseases. Currently, most methods for predicting pathogenic inframe indel mutations are based on machine learning, utilizing manually engineered features for prediction, which may overlook certain mutation data. Moreover, existing deep learning methods mainly focus on protein sequence features, neglecting other aspects. This paper proposes a method for predicting pathogenic inframe indel mutations, termed DPPred-indel, based on a combination of a biological language model and feature fusion. DPPred-indel leverages transfer learning to address the limited data availability and integrates features from both protein and DNA sequence levels for prediction. Subsequent experiments validate the effectiveness of transfer learning and feature fusion. DPPred-indel demonstrates comparable performance to state-of-the-art methods on two independent test sets constructed in this study.