ProtBERT-Based Prediction of Functional TRPV Sequences for Oral Cancer Pain

Abstract

Background and Objectives: Oral Cancer often leads to opioid tolerance, affecting pain-sensing neurons. TRPV ion channels, which are activated by inflammatory molecules, can be agonists or antagonists. Cancer changes the peripheral and central nervous systems, affecting tumor interactions, nociceptors, and the immune system. Activation of these receptors triggers pain perception and second messengers. Antagonists alleviate cancer-induced pain, suggesting therapeutic approaches. This study uses a large language model to find and forecast functional TRPV sequences for oral cancer pain.

Materials and Methods: TRPV proteins were obtained from UniProt; id -Q8NER1, Q9HBA0, Q9NQA5, Q9H1D0, and Q9Y5S1 were identified and quality-checked. Deepbio analyzed FASTA sequences. Large language models and sequence prediction techniques were used for prediction and classification, including Protbert (a pre-trained protein sequence model), BiLSTM, LSTMAttention, and TextRGNN (a residual graph neural network).

Results: The study reveals that ProtBERT and BiLSTM are the best predictors of amino acid sequences for TRPV1-based oral cancer, offering a balance of sensitivity and specificity. The ProtBERT model effectively identifies positive and negative cases related to TRPV with an accuracy of 0.90, a sensitivity of 0.89, a specificity of 0.91, and an AUC of 0.937. The BiLSTM model, similar to ProtBERT, has high accuracy and sensitivity, while the LSTM Attention model is reliable in detecting true positive cases. The text RGNN model has an accuracy of 0.51.

Conclusions: Machine learning models like ProtBERT and BiLSTM can enhance the early detection and diagnosis of oral cancer pain, but challenges like data quality and validation persist.