BigMHC Training and Evaluation Data

All training and evaluation data used in the BigMHC study (https://doi.org/10.1101/2022.08.29.505690). All code is freely available at https://github.com/KarchinLab/bigmhc -------------------------------------------------------------------------------- CSV Columns mhc - MHC-I allele pep - peptide sequence if epitope data or mutated peptide sequence if neoepitope data tgt - target value of 1 (presented/immunogenic) or 0 (non-presented/non-immunogenic) manafest.csv columns also include wtp (wild-type peptide) and gene (the name of the mutated gene) pseudoseqs.csv columns include the MHC-I allele along with the index and amino acid of the aligned positions. All other columns are the outputs of MHC-I epitope presentation and immunogenicity predictors. -------------------------------------------------------------------------------- datasets.zip contains the curated training, validation, and testing datasets along with a summary of the number of negatives and positives for each allele in each dataset (summary.csv): - el_test.csv - epitope presentation evaluation data and all evaluated model predictions - el_train.csv - epitope presentation training data - el_val.csv - epitope presentation validation data - im_test.csv - immunogenicity transfer learning evaluation data and all model predictions - im_train.csv - immunogenicity transfer learning training data - im_val.csv - immunogenicity transfer learning validation data - iedb.csv - infectious disease epitope evaluation data and all model predictions - summary.csv - table of positives and negatives across each allele for each dataset - manafest.csv - neoepitope immunogenicity data validated using MANAFEST assays - pseudoseqs.csv - one-hot encoded MHC representations el.csv.zip contains the predictions of the BigMHC production models and all other methods on all EL data (train, val, test). The pMHCs were filtered so that all other methods can score them (e.g. peptide lengths 8-11). eltrainval_models.zip contains the the models used to evaluate BigMHC on el_test.csv (the production models can be found in the GitHub repository) -------------------------------------------------------------------------------- sha256 sums are below: datasets.zip - 0d152a452756cf2e0014ffced6afc25118e7c11cf1f626b26e49f50f79edffaa el.csv.zip - cb94b42406b96a3b13d941cf87dd43f4f53e9ebfbc3d1619f0e43327f1fb6395 eltrainval_models.zip - e8500173cb2afbe5f8e8c0ebc60785c0de6d91aab622f8b83fff3b4e65b43223 manafest.csv - accf19b8bb797ec842c3ee1b1ce1966feb35035df746f7a56b2994403ba1ad99 pseudoseqs.csv - cd1fa24fb4c9fc0ee592a3d753458c4e3abed0d5cc4ca76e325aa274df8e900a本数据集涵盖BigMHC研究（https://doi.org/10.1101/2022.08.29.505690）所使用的全部训练与评估数据。所有代码均可于https://github.com/KarchinLab/bigmhc 免费获取。 -------------------------------------------------------------------------------- CSV文件列说明： mhc：MHC-I等位基因（MHC-I allele） pep：若为表位数据集则对应肽序列，若为新表位数据集则对应突变肽序列 tgt：目标值，取值为1（代表已呈递/具有免疫原性）或0（代表未呈递/无免疫原性） manafest.csv 的列还包含wtp（野生型肽，wild-type peptide）与gene（突变基因的名称） pseudoseqs.csv 的列包含MHC-I等位基因，以及比对位点的索引与对应氨基酸。 其余所有列均为MHC-I表位呈递与免疫原性预测工具的输出结果。 -------------------------------------------------------------------------------- datasets.zip 包含经过整理的训练、验证与测试数据集，以及各数据集中每个等位基因的阳性、阴性样本数量汇总表（summary.csv）： - el_test.csv：表位呈递评估数据及所有已评估模型的预测结果 - el_train.csv：表位呈递训练数据 - el_val.csv：表位呈递验证数据 - im_test.csv：免疫原性迁移学习评估数据及所有模型预测结果 - im_train.csv：免疫原性迁移学习训练数据 - im_val.csv：免疫原性迁移学习验证数据 - iedb.csv：传染病表位评估数据及所有模型预测结果 - summary.csv：各数据集各等位基因的阳性、阴性样本统计表 - manafest.csv：经MANAFEST实验验证的新表位免疫原性数据 - pseudoseqs.csv：独热编码的MHC表征 -------------------------------------------------------------------------------- el.csv.zip 包含BigMHC量产模型与其他所有方法在全部EL数据集（训练、验证、测试集）上的预测结果。我们对肽-MHC复合物（pMHC）进行了筛选，以确保所有其他方法均可对其进行评分（例如肽长度为8-11个氨基酸）。 eltrainval_models.zip 包含用于在el_test.csv上评估BigMHC的模型（量产模型可于GitHub仓库中获取） -------------------------------------------------------------------------------- 以下为各文件的SHA256校验和： datasets.zip：0d152a452756cf2e0014ffced6afc25118e7c11cf1f626b26e49f50f79edffaa el.csv.zip：cb94b42406b96a3b13d941cf87dd43f4f53e9ebfbc3d1619f0e43327f1fb6395 eltrainval_models.zip：e8500173cb2afbe5f8e8c0ebc60785c0de6d91aab622f8b83fff3b4e65b43223 manafest.csv：accf19b8bb797ec842c3ee1b1ce1966feb35035df746f7a56b2994403ba1ad99 pseudoseqs.csv：cd1fa24fb4c9fc0ee592a3d753458c4e3abed0d5cc4ca76e325aa274df8e900a

搜集汇总

背景与挑战

背景概述

该数据集是BigMHC研究中用于训练和评估MHC-I表位呈递及免疫原性预测模型的核心数据，包含多个CSV文件，如训练、验证、测试集，以及新生抗原免疫原性数据（manafest.csv）和MHC编码表示（pseudoseqs.csv）。其特点在于提供了详细的肽序列、等位基因和目标值（1表示呈递/免疫原性，0表示非呈递/非免疫原性），并整合了多种预测模型的输出，适用于免疫学领域的机器学习应用。