基于深度学习的预测TCR和肽的相互作用算法研究

Research on Predicting TCR and Peptide Interaction Algorithms Based on Deep Learning

摘要: 免疫疗法是一种利用人体自身免疫系统反应的癌症治疗方法。继手术、化疗和放疗等传统治疗方法后，免疫疗法逐渐成为了最有前途的癌症治疗方法。在各种免疫疗法中，发展较快的是TCR-T疗法。TCR-T疗法通过基因编辑技术，把特异性识别肿瘤抗原肽的T细胞受体(TCR)基因导入到患者自身的T细胞，使患者的T细胞能够表达外源性TCR，并且获得特异性杀伤肿瘤细胞的能力。然而，如何从大量非结合TCR中筛选出特异性识别肿瘤抗原肽的TCR，这一直是免疫和生物信息学领域的挑战。为解决这一问题，本文提出了一种基于深度学习的分类算法，对TCR与多肽的相互作用进行准确预测。实验结果表明，本文提出算法在数据集上AUC的值为0.8513。此外，还与随机森林和NetTCR分类模型进行不同的分类指标对比，该算法的指标值都有着较高的提升。

Abstract: Immunotherapy is a type of cancer treatment that exploits the body’s own immune system response. Immunotherapy is emerging as the most promising cancer treatment after traditional treatments such as surgery, chemotherapy and radiotherapy. Among the various immunotherapies, TCR-T therapy develops rapidly. TCR-T therapy introduces the T Cell Receptor (TCR) gene that specifically recognizes tumor antigen peptide into the patient’s T cells by gene editing technology, so that the patient’s T cells can express exogenous TCR and obtain the ability to specifically kill tumor cells. However, how to screen out TCRs that specifically recognize tumor antigenic peptides from a large number of unbound TCRs has been a challenge in the fields of immunity and bioinformatics. To solve this problem, this paper proposes a classification algorithm based on deep learning to accurately predict the interaction between TCR and peptides. The experimental results show that this paper proposes that the algorithm has a value of AUC of 0.8513 on the dataset. In addition, there are also different classification indexes compared with the random forest and NetTCR classification models, and the index value of this algorithm is highly improved.