Data for: ATR-FTIR Spectral Fingerprinting of Serum with Machine Learning Enables Single-Measurement Diagnosis and Prognostic Risk Assessment for Sepsis
收藏资源简介：
README: ML-Assisted Serum Fingerprinting for Sepsis Title: ATR-FTIR Spectral Fingerprinting of Serum with Machine Learning Enables Single-Measurement Diagnosis and Prognostic Risk Assessment for SepsisAuthors: Xuejie Wang, Wei Xin, Tangdong Chen, et al. --------------------------------------------------1. OVERVIEW--------------------------------------------------This dataset provides a complete evidence chain for the diagnosis and prognosis of sepsis using ATR-FTIR spectroscopy and machine learning. It includes raw spectra, preprocessed data, final feature matrix, and the computational codebase required for reproducibility.--------------------------------------------------2. FILE DESCRIPTIONS-------------------------------------------------- FILE 1: Raw_Data_Full (Folder/Zip)- Description: Original, unprocessed raw absorbance spectra.- Spectral Range: 4000–400 cm⁻¹- Status: No mathematical treatment applied. FILE 2: Processed_Dataset.CSV- Description: The preprocessed spectral matrix used for computational analysis and visualization (e.g., Figure 1).- Applied Preprocessing: 1. Rubber band baseline correction. 2. Savitzky–Golay smoothing (2nd order polynomial, 9-point window). 3. Min-Max normalization based on the Amide I band (approx. 1636 cm⁻¹). 4. -Data Range: Truncated to 4000–850 cm⁻¹- Structure: * Column 1: Sample_ID * Column 2: Group (Sepsis or Control) * Remaining Columns: Processed absorbance values (a.u.) per wavenumber.- Sample Size: n = 146 (including 78 sepsis patients and 68 controls). FILE 3: Feature_Vector_Table.CSV- Description: The final input table for Machine Learning (ML) algorithms (Extra Trees, Random Forest, etc.).- Content: Integrated area values of the six characteristic regions: Region 1: 2997–2887 cm⁻¹ Region 2: 1721–1589 cm⁻¹ Region 3: 1589–1478 cm⁻¹ Region 4: 1275–1203 cm⁻¹ Region 5: 1184–1140 cm⁻¹ Region 6: 1140–956 cm⁻¹- Purpose: Supports results in Figures 2-4. FILE 4: Clinical_Metadata.CSV- Description: Clinical characteristics and 28-day survival outcomes for the sepsis patients.- Sample Size: n = 83 (Patient sepsis-66 is excluded due to missing records).- Purpose: Supports the baseline characteristics (Table 4) and the prognostic scoring system (Figure 3).- Structure: * Column 1: Sample_ID * Column 2: 28-day survival outcomes * Remaining Columns: Clinical markers (e.g., PLT, PCT, CRP, etc.) used for correlation and ROC analysis. --------------------------------------------------3. COMPUTATIONAL SCRIPTS & RESULTS (CODE)--------------------------------------------------This section provides the Python codebase and detailed source data results. FILE 5: Code_and_Results.zip│├── script_diagnosis.py # Python script for Sepsis vs. Control classification.├── script_prognosis.py # Python script for survival outcome prediction.│├── /Data/ # Input datasets for machine learning scripts.│ ├── control.csv # Preprocessed spectral data for healthy controls.│ ├── sepsis.csv # Preprocessed spectral data for sepsis patients.│ └── prognosis_data.csv # Combined clinical and spectral data for prognosis.│└── /Results/ # Exhaustive model output for reproducibility. ├── source_data_diagnosis_results.xlsx │ # Metrics: Performance comparison, Feature Importance, Confusion Matrix, │ # and ROC curve data for all 16 tested algorithms. └── source_data_prognosis_results.xlsx # Metrics: Comparison of 10 prognostic models, Naive Bayes details, # and Precision-Recall (PR) data. --------------------------------------------------4. SOFTWARE & ENVIRONMENT--------------------------------------------------- Spectral Acquisition: Nicolet iS50 FTIR spectrometer (Thermo Fisher Scientific).- Preprocessing Software: OMNIC 9.2 and Origin 2021.- ML Implementation: Python 3.10.0 (Scikit-learn, Pandas, Numpy, Matplotlib).Note on Environment: A requirements.txt file is provided within the Code_and_Results.zip. To install all necessary dependencies, run pip install -r requirements.txt. --------------------------------------------------5. CONTACT--------------------------------------------------For questions regarding data or code, please contact:Guoqiang Bao (guoqiang@fmmu.edu.cn) or Lijuan Yuan (lijuanyuan@fmmu.edu.cn).
README：用于脓毒症的机器学习辅助血清指纹图谱分析 标题：基于机器学习的血清衰减全反射傅里叶变换红外光谱（ATR-FTIR）指纹图谱可实现脓毒症单次检测诊断与预后风险评估 作者：王学杰、辛伟、陈唐东等 --------------------------------------------------1. 概述-------------------------------------------------- 本数据集提供了利用衰减全反射傅里叶变换红外光谱（ATR-FTIR）与机器学习（Machine Learning, ML）开展脓毒症诊断与预后研究的完整证据链，包含原始光谱数据、预处理后的数据、最终特征矩阵以及可复现研究所需的计算代码库。 --------------------------------------------------2. 文件说明-------------------------------------------------- 文件1：Raw_Data_Full（文件夹/压缩包） 描述：原始未处理的吸光度光谱数据。 光谱范围：4000–400 cm⁻¹ 状态：未施加任何数学处理。 文件2：Processed_Dataset.CSV 描述：用于计算分析与可视化（如图1）的预处理光谱矩阵。 所采用的预处理步骤如下： 1. 橡胶带基线校正； 2. 萨维茨基-戈莱平滑（2阶多项式，9点窗口）； 3. 基于酰胺I带（约1636 cm⁻¹）的最小-最大归一化； 4. 数据范围截断至4000–850 cm⁻¹ 结构： * 第1列：Sample_ID（样本编号） * 第2列：分组（脓毒症组或对照组） * 其余列：各波数下的吸光度值（任意单位，a.u.） 样本量：n = 146（含78名脓毒症患者与68名健康对照者）。 文件3：Feature_Vector_Table.CSV 描述：机器学习（ML）算法（极端随机树（Extra Trees）、随机森林（Random Forest）等）的最终输入表。 内容为6个特征区域的积分面积值： 区域1：2997–2887 cm⁻¹ 区域2：1721–1589 cm⁻¹ 区域3：1589–1478 cm⁻¹ 区域4：1275–1203 cm⁻¹ 区域5：1184–1140 cm⁻¹ 区域6：1140–956 cm⁻¹ 用途：支撑图2-4的研究结果。 文件4：Clinical_Metadata.CSV 描述：脓毒症患者的临床特征与28天生存结局数据。 样本量：n = 83（因记录缺失排除样本sepsis-66）。 用途：支撑基线特征表（表4）与预后评分系统（图3）。 结构： * 第1列：Sample_ID（样本编号） * 第2列：28天生存结局 * 其余列：用于相关性分析与ROC曲线分析的临床标志物（如血小板PLT、降钙素原PCT、C反应蛋白CRP等）。 --------------------------------------------------3. 计算脚本与结果（代码）-------------------------------------------------- 本节提供Python代码库与详细的源数据结果。 文件5：Code_and_Results.zip ├── script_diagnosis.py # 用于脓毒症与对照组分类的Python脚本 ├── script_prognosis.py # 用于生存结局预测的Python脚本 ├── /Data/ # 机器学习脚本的输入数据集 │ ├── control.csv # 健康对照者的预处理光谱数据 │ ├── sepsis.csv # 脓毒症患者的预处理光谱数据 │ └── prognosis_data.csv # 用于预后分析的临床与光谱整合数据 └── /Results/ # 用于复现研究的完整模型输出 ├── source_data_diagnosis_results.xlsx │ # 指标：16种测试算法的性能对比、特征重要性、混淆矩阵、ROC曲线数据 └── source_data_prognosis_results.xlsx # 指标：10种预后模型的性能对比、朴素贝叶斯模型细节、精确率-召回率（PR）曲线数据 --------------------------------------------------4. 软件与环境-------------------------------------------------- - 光谱采集设备：Nicolet iS50傅里叶变换红外光谱仪（赛默飞世尔科技，Thermo Fisher Scientific） - 预处理软件：OMNIC 9.2与Origin 2021 - 机器学习实现环境：Python 3.10.0（依赖Scikit-learn、Pandas、Numpy、Matplotlib库） 环境说明：Code_and_Results.zip压缩包内附带requirements.txt文件，可通过运行`pip install -r requirements.txt`安装所有必要依赖。 --------------------------------------------------5. 联系方式-------------------------------------------------- 若对数据或代码有疑问，请联系： 鲍国强（guoqiang@fmmu.edu.cn）或袁丽娟（lijuanyuan@fmmu.edu.cn）
