肺癌是癌症死亡的主要原因。随着靶向治疗的发展，肺癌患者的预后和生存有很大的改善，但仍存在部分患者对治疗耐药。CT-707是一种口服二代间变性淋巴瘤激酶（Anaplastic lymphoma kinase，ALK）酪氨酸激酶抑制剂（Tyrosine kinase inhibitor，TKI）。鉴于肺癌患者接受CT-707治疗的疗效异质性，探索与CT-707疗效相关的因素，构建疗效预测模型，将有助于寻找CT-707获益患者。此外，由于SARS-CoV-2疫情爆发，SARS-CoV-2疫苗被研发和广泛接种，而大多数IV期肺癌患者，需同时接受靶向治疗等抗肿瘤治疗。因此评估肺癌患者接种灭活SARS-CoV-2疫苗以及Omicron感染后的免疫反应并探讨抗肿瘤靶向治疗等因素对免疫反应的影响具有重要意义。本论文的研究内容包含两个部分：第一部分以接受二代ALK TKI CT-707治疗的ALK阳性非小细胞肺癌（Non-small cell lung cancer，NSCLC）患者为研究对象，利用临床药代动力学和常规临床指标数据，构建疗效预测模型，并开发面向研究人员的界面化系统，直观展现疗效预测结果。同时利用人蛋白芯片，筛选并探索肿瘤自身抗体标志物的疗效预测价值，弥补临床实验室常规检测指标的不足。第二部分以肺癌患者为研究对象，探究SARS-CoV-2疫苗在肺癌患者的覆盖率、免疫原性和对Omicron感染后的保护性，探讨抗肿瘤靶向治疗等因素对SARS-CoV-2疫苗和Omicron感染诱导免疫反应的影响。

第一部分 ALK阳性非小细胞肺癌患者接受靶向治疗疗效预测和标志物研究

ALK重排的NSCLC患者预后较差，累及中枢神经系统的风险较高。克唑替尼是第一代ALK抑制剂，然而大多数经克唑替尼治疗的患者最终出现获得性耐药。CT-707是一种ATP竞争性的二代ALK TKI。由于CT-707 I期-III期临床试验中的NSCLC患者接受CT-707治疗的疗效存在差异，有必要深入探究影响CT-707疗效的因素，探索CT-707治疗的疗效预测相关血浆生物标志物，构建二代ALK TKI CT-707疗效预测模型，以便于患者分层和寻找二代ALK TKI获益患者。

针对以上问题，我们收集348例来自CT-707 I期、II期和III期临床试验的NSCLC患者的基线临床数据和药代动力学等，根据患者最佳整体疗效，将其分为有疗效组和无疗效组。来自CT-707 I期的53例患者全部纳入训练集、来自II期既往接受克唑替尼治疗的115例患者随机划分为训练集（n＝60）和验证集（n＝55）、来自III期既往未接受克唑替尼治疗的180例患者随机划分为训练集（n＝90）和验证集（n＝90）。使用贪婪算法（Greedy algorithm）从训练集筛选用于模型构建的指标，采用混合专家模型（Mixture of experts，MoE）-多层感知器（Multi-layer perceptron，MLP）、MoE-残差神经网络（Residual network，ResNet）和MoE-密集连接卷积网络（Densely connected convolutional networks，DenseNet）三种深度学习算法，构建和验证临床预测模型。随后我们利用自身抗体蛋白芯片，检测CT-707 I期临床试验中51例NSCLC患者治疗前基线样本中的自身抗体，并用ELISA方法测定来自CT-707 II期和III期临床试验的训练集150例和验证集145例NSCLC基线样本中自身抗体表达，同时结合患者临床指标构建和验证临床-自身抗体联合预测模型。

本研究首先纳入训练集203例，验证集145例。贪婪算法结果显示训练集中预测患者接受CT-707治疗疗效的最佳组合为克唑替尼治疗史、ECOG PS、稳态谷浓度和白蛋白，F1分数（F1 Score）＝0.7710，将其用于模型构建和验证。MoE-MLP模型在训练集和验证集的曲线下面积（Area under curve，AUC）分别为0.8353、0.7324，较MoE-ResNet（训练集AUC＝0.6818，验证集AUC＝0.6157）和MoE-DenseNet（训练集AUC＝0.6982，验证集AUC＝0.6332）诊断性能高。此外，我们基于MoE-MLP模型开发了一个面向研究人员的界面化系统，实现研究者前端输入受试者参数后，后台自动根据模型参数进行分析，可直观展示患者接受CT-707治疗的预期疗效。

将51例NSCLC患者自身抗体检测结果与患者临床检测指标结合，进行随机支持向量（Support vactor machine，SVM）分析，发现有克唑替尼治疗史、NPHS2自身抗体、单核细胞计数、APTT、MRPS21自身抗体和COL21A1自身抗体是预测肺癌患者CT-707治疗疗效的重要指标。将上述3种自身抗体与上述3个临床指标结合，构建并验证临床-自身抗体联合MoE-MLP模型（训练集AUC＝0.7161，验证集AUC＝0.7061），模型性能优于仅纳入临床指标（训练集AUC＝0.6923，验证集AUC＝0.6852）和仅纳入自身抗体指标（训练集AUC＝0.6017，验证集AUC＝0.5868）的模型。

以上研究证明了深度学习在临床试验中的适用性，同时这种方法可以有效地发现和验证影响疗效的因素。自身抗体对NSCLC患者接受二代ALK TKI靶向治疗疗效有预测价值，可用于弥补临床实验室常规检测指标的不足。

第二部分 肺癌患者新型冠状病毒疫苗接种免疫原性和保护性研究

截至2023年8月16日，严重急性呼吸系统综合征冠状病毒2（SARS-CoV-2）及其变异株已导致全球大流行，全球有7.69亿多例病例和695万人死亡。由于Omicron变异株的高传播性，2022年12月至2023年2月7日，超过82%的中国人口在两个月内被感染。疫苗接种一直是预防病毒感染的基础。然而，SARS-CoV-2疫苗在中国大陆肺癌患者接种率、安全性、免疫反应和预防Omicron重复感染的有效性尚不清楚。

因此，为解决以上问题，我们在2022年10月至2022年11月期间对1018例肺癌患者SARS-CoV-2疫苗接种率以及影响接种意愿的因素开展了电子问卷调研，并对260例肺癌患者、140例健康对照和另外40例有连续采血点的肺癌患者SARS-CoV-2疫苗接种后的免疫反应进行横断面研究，使用ELISA法检测肺癌患者接种第二剂和加强剂灭活SARS-CoV-2疫苗后14-90天、91-180天和＞180天SARS-CoV-2特异性总抗体、抗RBD IgG抗体，SARS-CoV-2野生型（Wild type，WT）和Omicron BA.4/5变异株中和抗体水平，并通过多因素回归分析影响SARS-CoV-2特异性抗体阳转的因素。随后，于2023年3月至2023年8月在447例肺癌患者开展了Omicron感染后的免疫反应纵向研究，通过ELISA检测肺癌患者Omicron感染后3个月和7个月的上述SARS-CoV-2特异性抗体水平，同时采用酶联免疫斑点技术（Enzyme Linked Immunospot Assay, Elispot）检测肺癌患者Omicron感染后3个月的抗SARS-CoV-2 WT和Omicron变异株的T细胞反应，并对7个月内肺癌患者Omicron二次感染情况进行随访，探究SARS-CoV-2疫苗对肺癌患者Omicron二次感染的保护性。

在1018例肺癌患者中，549例（54%）接种过SARS-CoV-2疫苗，有75例（13.7%）患者报告了可接受的全身不良事件，其中最常见的是发热（7%）。女性（OR＝1.512，95% CI＝1.076-2.124），城市居民（OR＝2.048，95% CI＝1.238-3.389），正在接受治疗（OR＝2.897，95% CI＝1.348-6.226），怀疑疫苗安全性（OR＝3.816，95% CI＝2.198-6.626）是疫苗犹豫的独立风险因素。在373名已接种第三剂疫苗的肺癌患者中，仅167例（44.8%）愿意接受第四剂SARS-CoV-2疫苗，主要原因是担心第四剂疫苗会影响肺癌进展和抗癌治疗或有相关的副作用。

灭活疫苗加强剂增强了肺癌患者SARS-CoV-2特异性体液免疫反应，包括抗SARS-CoV-2总抗体、抗RBD IgG抗体、抗SARS-CoV-2 WT中和抗体和Omicron BA.4/5的中和抗体，但抗体反应较健康人低，且增强的体液免疫反应随着时间的推移而减弱，尤其是WT（14-90天 vs. ＞180天，P＝0.0035）和BA.4/5中和抗体（14-90天 vs. ＞180天，P＝0.0002）。肺癌患者加强免疫后14-90天中和抗体对BA.4/5的抑制率极低（18.82%），加强免疫未能诱导肺癌患者产生针对Omicron BA.4/5强而持久的免疫应答。年龄≥65岁是WT中和抗体阳转的危险因素（OR＝0.322，95% CI＝0.134-0.773，P＝0.0112）。接受抗肺癌治疗（化疗、靶向治疗、免疫治疗和放疗）的患者抗体反应低于未接受系统治疗的患者，放疗是WT中和抗体阳转的危险因素（OR＝0.082，95% CI＝0.011-0.617，P＝0.0151）。肺癌患者体液免疫低下可能是由于化疗或放疗等诱导的淋巴细胞减少，可能与总B细胞、CD4⁺ T细胞和CD8⁺ T细胞计数降低有关。

肺癌患者感染Omicron后症状较轻，SARS-CoV-2疫苗对肺癌患者发热有一定疗效，而SARS-CoV-2特异性抗体水平与长新冠症状无关。Omicron感染诱导肺癌患者SARS-CoV-2特异性抗体水平显著增加，完全免疫或加强免疫的肺癌患者体液免疫应答高于未接种者，免疫反应强度至少持续7个月。感染后肺癌患者对Omicron特异性表位的T细胞应答强度高于WT。9.74%肺癌患者在随访期间发生了二次感染，其SARS-CoV-2特异性抗体水平低于未二次感染患者，而未接种疫苗或未全程接种疫苗，仅接受免疫治疗的晚期肺癌患者是二次感染的高危人群。

以上研究表明了新冠疫苗在肺癌人群的安全性和有效性，及其对二次感染的保护性，为肺癌患者的新冠疫苗接种政策提供了数据支持。

Lung cancer is the leading cause of cancer death. With the development of targeted therapy, the prognosis and survival of lung cancer patients have been greatly improved, but there are still some patients resistant to treatment. CT-707 is an oral second-generation anaplastic lymphoma kinase (ALK) tyrosine kinase inhibitor (TKI). Given the heterogeneity of the efficacy of CT-707 treatment in patients with lung cancer, it is of great value to explore the factors related to the efficacy of CT-707 and construct a predictive model for the efficacy of CT-707 to find the patients who benefit from CT-707. In addition, due to the outbreak of the SARS-CoV-2 epidemic, the SARS-CoV-2 vaccine has been developed and widely inoculated. However, most patients with stage IV lung cancer need to continue to receive anti-tumor treatment such as targeted therapy or chemotherapy. Therefore, it is of great significance to evaluate the immune response of lung cancer patients after vaccination with inactivated SARS-CoV-2 vaccine and Omicron infection and to explore the influence of treatment and other factors on the immune response. This research consists of two parts: In the first part, ALK-positive non-small cell lung cancer (NSCLC) patients treated with CT-707 were the subjects. The clinical pharmacokinetic and routine clinical data were used to construct the efficacy prediction model, and the interactive tool for researchers was developed so that the efficacy prediction results could be presented intuitively. At the same time, the human protein chip platform was used to explore the predictive value of the autoantibodies and to make up for the deficiency of routine clinical laboratory indicators. In the second part, lung cancer patients were taken as the research object to explore the coverage, immunogenicity, and protection of the SARS-CoV-2 vaccine after Omicron infection in lung cancer patients, and to explore the influence of anti-tumor treatment such as targeted therapy and other factors on SARS-CoV-2 vaccine and Omicron induced immune response.

Section I Prediction of efficacy and biomarkers of targeted therapy in ALK positive NSCLC patients

The prognosis of NSCLC patients with ALK rearrangement is poor, and the risk of central nervous system involvement is high. Crizotinib is the first-generation ALK inhibitor. However, most patients treated with crizotinib eventually develop acquired resistance. CT-707 is an ATP-competitive second-generation ALK TKI. Given the heterogeneity of the efficacy of CT-707 treatment in patients with ALK-positive NSCLC, it is of great necessity to find the factors related to the efficacy of patients from CT-707 phase I to phase III clinical trials, explore the efficacy-related plasma biomarkers of CT-707 treatment, and construct the efficacy prediction model for facilitating patients stratification and finding patients who could benefit from second-generation ALK TKI.

To resolve the above problems, we collected baseline clinical data and pharmacokinetics of 348 NSCLC patients from CT-707 phase I, phase II and phase III clinical trials. According to the best overall efficacy of patients, they were divided into effective group and non-effective group. All 53 patients from CT-707 phase I were included in the training set, 115 patients from phase II who had received crizotinib treatment were randomly divided into the training set (n = 60) and the validation set (n = 55), and 180 patients from phase III who had not received crizotinib treatment were randomly divided into the training set (n = 90) and the validation set (n = 90). Greedy algorithm was used to screen the indicators for model construction from the training set. Three deep learning algorithms, Mixture of experts (MoE) -Multilayer perceptron (MLP), MoE-Residual network (ResNet) and MoE-Densely connected convolutional networks (DenseNet), were used to construct and verify the clinical prediction model for predicting the efficacy of ALK-positive NSCLC patients after CT-707 treatment.

Then we used focused microprotein assay to detect the autoantibodies in the baseline samples of 51 NSCLC patients in the CT-707 phase I clinical trial and combined them with the clinical indicators for feature selection. The expression of autoantibodies was detected by ELISA in the baseline samples of 150 NSCLC patients in the training set and 145 NSCLC patients in the validation set from CT-707 phase II and phase III clinical trial. At the same time, combined with the clinical indicators of patients, the clinical-autoantibody combined prediction model was constructed and verified.

In this study, the training set and validation set were divided into 203 patients and 145 patients. The results of the greedy algorithm showed that the best combination of predictors for the efficacy of patients receiving CT-707 in the training set was the crizotinib usage history, ECOG PS, steady-state trough concentration (C_{ss_min}) and albumin (F1 = 0.7710), which was used for model construction and verification. The area under curve (AUC) of the MoE-MLP model in the training set and the validation set were 0.8353 and 0.7324, respectively, which were higher than that of MoE-ResNet (training set AUC = 0.6818, validation set AUC = 0.6157) and MoE-DenseNet (training set AUC = 0.6982, validation set AUC = 0.6332). In addition, we developed an interface system for researchers based on the MoE-MLP model. After the researchers input the subject parameters, the system automatically performs statistical analysis based on the model parameters established by the training set data, which could intuitively show the expected efficacy of patients receiving CT-707, which is helpful for the stratification of efficacy and the management of clinical trials by the principal investigator.

The baseline samples of 51 NSCLC patients before treatment were tested for autoantibodies, and the clinical indicators of patients were analyzed by support vactor machine (SVM). It was found that the crizotinib therapy history, monocyte count, APTT, NPHS2 autoantibodies, MRPS21 autoantibodies, and COL21A1 autoantibodies were of great value in predicting the efficacy of CT-707 treatment in patients with lung cancer. The ELISA platform was used to detect these three autoantibodies in the baseline samples of 150 NSCLC patients in the training set and 145 NSCLC patients in the validation set. Combining these three autoantibodies with the above three clinical indicators, a clinical-autoantibody combined MoE-MLP model for clinical efficacy and prognosis was constructed and validated (training set AUC = 0.7161, validation set AUC = 0.7061). The performance of the model was better than that of only clinical indicators (training set AUC = 0.6923, validation set AUC = 0.6852) and inclusion of autoantibodies (training set AUC = 0.6017, validation set AUC = 0.5868).

The above research proved the applicability of deep learning in clinical trials, and this method can effectively discover and verify the factors affecting the efficacy by analyzing multiple heterogeneous patient-related variables. Autoantibodies have predictive value for the efficacy of second-generation ALK TKI, which can be used to make up for the deficiency of routine detection indicators in clinical laboratories.

Section II Immunogenicity and protection of SARS-CoV-2 vaccines in patients with lung cancer

As of August 16,2023, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused global pandemic, with more than 769 million cases and 6.95 million deaths worldwide. Due to the high transmissibility of the Omicron variant, between December 2022 and February 7, 2023, more than 82% of the population was infected within two months. However, the vaccination rate, safety, immune response and effectiveness of preventing Omicron reinfection of SARS-CoV-2 vaccine in lung cancer patients in mainland China were still unclear.

Therefore, to solve the above problems, we conducted an electronic questionnaire survey on the vaccination uptake rate of the SARS-CoV-2 vaccine and the factors affecting the vaccination willingness of 1018 lung cancer patients from October, 2022 to November, 2022. A cross-sectional study was conducted on the immunogenicity of the SARS-CoV-2 vaccine in 260 lung cancer patients, 140 healthy controls and 40 lung cancer patients with continuous blood collection points. The specific total antibody of SARS-CoV-2, anti-RBD IgG antibody, the neutralizing antibodies (Nabs) of SARS-CoV-2 wild type (WT) and Omicron BA.4/5 strains were detected by ELISA in lung cancer patients 14-90 days, 91-180 days and > 180 days after inoculation with the second dose and enhancer inactivated SARS-CoV-2 vaccine. Logistic regression analysis was used to analyze the influencing factors of positive conversion of SARS-CoV-2 specific antibody. Thereafter, a longitudinal study of the immune response after Omicron infection in 447 patients from March 2023 to August 2023 were conducted. The above SARS-CoV-2 specific antibodies in lung cancer patients at 3 months and 7 months after Omicron infection were detected by ELISA, and the T cell responses against SARS-CoV-2 WT and Omicron strains in lung cancer patients at 3 months after Omicron infection were detected via enzyme linked immunospot assay (Elispot). The reinfection of Omicron in lung cancer patients within 7 months was followed up. To explore the protective effect of mixed immune response on the reinfection of Omicron in lung cancer patients.

Among the 1018 patients, 549 (54%) received SARS-CoV-2 vaccine, and 75 (13.7%) reported acceptable systemic adverse events, the most common of which was fever (7%). The female (OR = 1.512, 95% CI = 1.076-2.124), urban residents (OR = 2.048, 95% CI = 1.238-3.389), undergoing treatment (OR = 2.897, 95% CI = 1.348-6.226), and doubting vaccine safety (OR = 3.816, 95% CI = 2.198-6.626) were independent risk factors for vaccine hesitation. Of the 373 patients who received three doses of treatment, only 167 cases (44.8%) were willing to accept a fourth dose of treatment due to concerns about the safety and efficacy of the vaccine variant.

The booster of inactivated vaccine enhanced the SARS-CoV-2 specific antibody response in lung cancer patients, including anti-SARS-CoV-2 total antibody, anti-RBD IgG antibody, anti-SARS-CoV-2 WT Nabs and Omicron BA.4/5 Nabs, but the expression levels in lung cancer patients was lower than that in healthy people. After strengthening agents, the enhanced humoral response weakened over time, especially WT Nabs (14-90 days vs. > 180 days, P = 0.0035) and BA.4/5 Nabs (14-90 days vs. > 180 days, P = 0.0002). The inhibition rate of Nabs to BA.4/5 was very low (18.82%) at 14-90 days after booster immunization in lung cancer patients, so the booster immunization failed to induce a strong and lasting immune response against Omicron BA.4/5 in lung cancer patients. Age ≥ 65 was a risk factor for WT Nabs serum conversion (OR = 0.322, 95% CI = 0.134-0.773, P = 0.0112). The antibody responses of patients receiving anti-lung cancer treatment (chemotherapy, targeted therapy, immunotherapy, and radiotherapy) was lower than that of patients without treatment, and radiotherapy was a risk factor for WT Nabs positive conversion (OR = 0.082，95% CI = 0.011-0.617，P = 0.0151). The decrease of humoral immunity in patients with lung cancer may be due to the decrease of lymphocytes induced by chemotherapy or radiotherapy, which may be related to the decrement of total B cells, CD4⁺ T cells and CD8⁺ T cells.

The symptoms of lung cancer patients infected with Omicron were mild, and the vaccine had efficacy against fever of lung cancer patients. However, the level of SARS-CoV-2 specific antibodies were not related to the long novel coronary symptoms. Omicron infection induced a significant increase in SARS-CoV-2 specific antibody levels in lung cancer patients. The humoral response of lung cancer patients with complete or booster immunization was higher than that of unvaccinated patients, and the immune response lasted for at least 7 months. The response of T cells to Omicron was higher than that of WT in lung cancer patients after infection. In addition, 9.74% lung cancer patients had reinfection, and their antibody response level was lower than that of patients without reinfection, while those unvaccinated or not fully vaccinated, and only received anti-tumor immunotherapy late stage patients were at high risk of reinfection.

The above studies have shown the safety and effectiveness of the SARS-CoV-2 vaccine in the lung cancer population, and its protection against repeated infections, providing data support for the SARS-CoV-2 vaccine policy for lung cancer patients.