Novel Machine Learning Methods for Analysis of MALDI-TOF Mass Spectrometry Data

描述（由申请人提供）： 肝细胞癌（HCC）是一种常见的癌症，每年有多达50万例新发病例。1981年至1998年，HCC患者的5年生存率仅从2%上升到5%。这种较差的生存率部分与晚期HCC的诊断有关，其中缺乏有效的治疗方法。早期发现HCC可提高患者生存率。肝硬化患者通常会发展为HCC。因此，监测癌症患者可能会降低癌症相关的死亡率。目前可用的工具灵敏度和特异性差，阻碍了HCC监测的广泛实施。因此，需要额外的血清标志物，提供更高的灵敏度和特异性，以提高早期肝癌的检出率。该合作项目的目标是确定一组用于早期诊断HCC的血清生物标志物。长期目标是找到并验证有助于在高风险人群中识别可治疗阶段HCC的标志物。该项目将导致开发创新的质谱数据预处理和生物标志物选择方法，用于通过使用低分子量（LMW）富集血清的基质辅助激光解吸/电离飞行时间（MALDI-TOF）质谱（MS）鉴定HCC特异性的候选生物标志物。该项目的具体目标如下：目标1：开发用于改进MALDI-TOF质谱数据预处理的算法，包括离群值筛选、分箱、平滑、基线校正、归一化、峰检测和峰校准。所提出的算法将使我们能够减少标准血清重复谱的运行间变异性，并提高区分HCC患者和正常患者或健康个体的预测准确性。目标二：通过结合两种流行的机器学习方法，粒子群优化（PSO）和支持向量机（SVMs），开发一种新的算法，该算法上级于目前使用的生物标志物选择方法。所提出的算法将用于从预处理的MALDI-TOF光谱中识别HCC特异性标志物。为避免混淆效应，如果峰与病毒感染或协变量（如年龄、性别、吸烟状况、饮酒状况和居住地（城市或农村））相关，则在生物标志物选择前将其删除。从剩余的峰中，将鉴定出一小组准确区分HCC患者和HCC患者的候选生物标志物。该算法识别具有高灵敏度和特异性的一小组标记物的能力对于建立临床测试至关重要。此外，该算法将识别区分各种对（正常与肝硬化，正常与HCC，肝硬化与早期HCC，以及肝硬化与晚期HCC）的标志物。这将使我们能够分离HCC特异性标志物并鉴定疾病进展标志物。此外，将鉴定由所选候选生物标志物代表的肽。最后，将该算法的性能与现有的方法进行比较。

DESCRIPTION (provided by applicant): Hepatocellular carcinoma (HCC) is a common cancer worldwide with as many as 500,000 new cases each year. Between 1981 to 1998, the 5-year patient survival rate with HCC only rose from 2% to 5%. This poor survival rate is in part related to the diagnosis of HCC at advanced stages, where effective therapies are lacking. Early detection of HCC improves patient survival. Patients with cirrhosis are typically the ones to develop HCC. Hence, monitoring cirrhotic patients can potentially decrease the cancer-related mortality rate. The poor sensitivity and specificity of currently available tools has prevented widespread implementation of HCC surveillance. Therefore, additional serum markers that provide higher sensitivity and specificity are needed to improve the detection rate of early HCC. The goal of this collaborative project is to identify a panel of serum biomarkers for early diagnosis of HCC. The long-term goal is to find and validate markers that would help identify HCC at a treatable stage in high-risk population of cirrhotic patients. This project will lead to the development of innovative mass spectral data preprocessing and biomarker selection methods that for the identification of candidate biomarkers specific to HCC by using matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry (MS) of low-molecular-weight (LMW) enriched sera. The specific aims of the project are the following: Aim 1: To develop algorithms for improved MALDI-TOF mass spectral data preprocessing including outlier screening, binning, smoothing, baseline correction, normalization, peak detection, and peak calibration. The proposed algorithms will enable us to reduce run-to-run variability in replicate spectra of a standard serum and to enhance the prediction accuracy in distinguishing HCC patients from cirrhotic patients or healthy individuals. Aim 2: To develop a novel algorithm that is superior to currently used biomarker selection methods by combining two popular machine learning methods, particle swarm optimization (PSO) and support vector machines (SVMs). The proposed algorithm will be used to identify HCC-specific markers from the preprocessed MALDI-TOF spectra. To avoid confounding effects, peaks will be removed prior to biomarker selection if they are associated with viral infection or covariates such as age, gender, smoking status, drinking status, and residency (urban or rural). From the remaining peaks, a small set of candidate biomarkers that accurately distinguishes HCC patients from cirrhotic patients will be identified. The capability of the algorithm to identify a small set of markers with high sensitivity and specificity is critical for establishment of clinical tests. Additionally, the algorithm will identify markers that distinguish various pairs (normal vs. cirrhosis, normal vs. HCC, cirrhosis vs. early-stage HCC, and cirrhosis vs. late-stage HCC). This will enable us to isolate HCC- specific markers and identify disease progression markers. Furthermore, the peptides represented by the selected candidate biomarkers will be identified. Finally, the performance of the algorithm will be compared with existing methods.