📄 中文摘要
针对转移性黑色素瘤的治疗，采用了新抗原特异性CAR-T细胞，这种细胞能够精准靶向肿瘤特异性突变并通过HLA-I分子呈现。尽管检查点抑制剂已改善生存率，但耐药性问题依然存在。当前的新抗原CAR设计面临配体亲和力不足（<10 nM）、抗原发现困难及制造复杂性等瓶颈。为了解决这些问题，研究提出了一种集成平台，该平台通过全外显子组测序和机器学习预测HLA结合，识别患者特异性的高置信度新表位，并通过训练卷积神经网络优化CAR scFv亲和力。
深度学习引导的亲和力优化新抗原CAR-T治疗转移性黑色素瘤
📄 English Summary
**Deep‑Learning‑Guided Affinity‑Optimized Neoantigen CAR‑T for Metastatic Melanoma**
This study presents a novel approach for treating metastatic melanoma using neoantigen-specific CAR-T cells that precisely target tumor-specific mutations presented on HLA-I molecules. While checkpoint inhibitors have improved survival rates, resistance remains a significant challenge. Current neoantigen CAR designs are hindered by limited ligand affinity (<10 nM), difficulties in antigen discovery, and manufacturing complexities. To address these issues, an integrated platform is proposed that identifies patient-specific high-confidence neoepitopes through whole-exome sequencing and machine learning prediction of HLA binding, and optimizes CAR scFv affinity by training a convolutional neural network.