Cell-free DNA coverage and fragmentomic profiles produced under various sequencing protocols

This repository contains 4 data sets designed for assessing the validity of domain adaptation algorithms for bias correction in cell-free DNA data (coverage or fragmentomic profiles). Samples processed under the same or similar preanalytical settings are grouped into so-called domains.Each data subset has its own peculiarities:The NIPT set contains 1126 profiles from 563 biological samples processed twice with different library preparation or sequencing methods. The 1126 profiles are split in 12 domains, and each profile from any domain is in pair with a profile from another domain. For example, any profile in domain D3a has been sequenced with the Illumina HiSeq 2000 platform, and originates from a sample that has also been sequenced with the Illumina NovaSeq 6000 platform, resulting in a second profile in domain D3b. Domain names containing the same digit differ by one or 2 preanalytical variable(s) (e.g., the sequencing platform in the case of D3x). The domains present in the NIPT data set are: D1a, D1b, D2a, D2b, D3a, D3b, D4a, D4b, D5a, D5b, D6a, D6b.The OV set contains ovarian carcinoma cases and controls from 2 domains. The 2 domains correspond to 2 different bioinformatics teams, therefore both library preparation and sequencing protocols differ. 64 of the samples have been processed by both teams, resulting in coverage profiles in both domains. The domains present in the OV data set are: D9, D10.The HEMA set contains 238 haematological cancer cases (Hodgkin lymphoma, diffuse large B cell lymphoma and multiple myeloma) and 242 controls in one domain, and only controls (257) in the second domain. The domains present in the HEMA data set are: D7, D8.The FRAG set contains paired-end sequencing data. While some of the samples were originally sequenced for methylation analysis, they have been used for fragmentomic analysis in the present context. The data set contains 74 female controls and 51 breast cancer cases prepared with the NEBNext Enzymatic Methyl-seq kit, as well as 57 female controls prepared with the KAPA HyperPrep kit.The metadata.csv file contains the details of each coverage profile, including data set, domain and category (e.g., healthy, Hodgkin lymphoma). When the profile is paired with a profile from another domain, meaning they originate from the same biological sample, the identifier of the corresponding profile is reported in the "Paired-With" column.For the NIPT, HEMA and OV data sets, each coverage profile was produced by counting, for each 10kb bin, the reads for which the alignment starts in this bin. The profiles were then smoothed by a running average (of size 100 bins). The standard deviation of the running window has been kept track of for each bin position. Each zip file contains all the coverage profiles from one domain, and each coverage profile is stored as a compressed tsv file with the following columns:"BINDEX" is the index of the bin, essentially the row number."CHR" is the chromosome on which the bin is located."MEAN" is the average normalized coverage within a window of 100 contiguous bins centered around current bin. Each bin has a size of 10kb."SD" is the standard deviation computed likewise.The data set also contains 4 supplementary files:gc-content-1000kb.csv: GC-content for each bin of the reference genome HG38. The number of lines in this file corresponds to the number of bins in the coverage profiles files after removing allosomal and mitochondrial bins. -1 corresponds to blacklisted regions.mappability-1000kb.csv: Mappability of each bin.blacklisted-10kb-bins.csv: List of blacklisted 10 kb bins.D11-D12-batches.json: Mapping between the sample names from the FRAG data set to the library preparation period (i.e., month). Sequencing batches are strictly included in library preparation periods. This information is used for leave-one-batch-out cross-validation to prevent batch effects from producing overoptimistic performance.<br>GC content and mappability information was taken from:<br>https://github.com/broadinstitute/ichorCNA/tree/master/inst/extdata本仓库包含4个数据集，用于评估无细胞DNA（cell-free DNA, cfDNA）数据中用于偏差校正的域自适应算法的有效性（覆盖度或片段组学特征）。在相同或相似分析前处理条件下得到的样本被划分为所谓的域（domain）。每个数据子集均有其独特性： NIPT数据集包含来自563份生物样本的1126组特征，这些样本采用不同的文库制备或测序方法完成了两次处理。1126组特征被划分为12个域，每个域中的特征均与另一域中的特征成对存在。例如，域D3a中的任意一组特征均采用Illumina HiSeq 2000平台测序，其来源样本同时使用Illumina NovaSeq 6000平台完成了测序，从而在域D3b中得到第二组特征。包含相同数字的域仅在1或2个分析前变量上存在差异（如D3x示例中的测序平台）。NIPT数据集包含的域为：D1a、D1b、D2a、D2b、D3a、D3b、D4a、D4b、D5a、D5b、D6a、D6b。 OV数据集包含来自2个域的卵巢癌病例与对照样本。这2个域对应2个不同的生物信息学团队，因此文库制备与测序流程均存在差异。其中64份样本由两个团队分别完成处理，从而在两个域中均得到覆盖度特征。OV数据集包含的域为：D9、D10。 HEMA数据集包含238例血液系统恶性肿瘤病例（霍奇金淋巴瘤、弥漫大B细胞淋巴瘤与多发性骨髓瘤）与242例对照样本（位于单个域中），以及仅包含257例对照样本的第二个域。HEMA数据集包含的域为：D7、D8。 FRAG数据集包含双端测序数据。尽管部分样本最初是为甲基化分析进行测序，但在本研究中被用于片段组学分析。该数据集包含74例女性对照与51例乳腺癌病例，均使用NEBNext Enzymatic Methyl-seq试剂盒完成文库制备，同时包含57例使用KAPA HyperPrep试剂盒制备文库的女性对照样本。 metadata.csv文件包含每份覆盖度特征的详细信息，包括所属数据集、域与类别（如健康对照、霍奇金淋巴瘤）。若某一特征与另一域中的特征成对存在（即二者来源于同一份生物样本），则"Paired-With"列中会标注对应特征的标识符。 针对NIPT、HEMA与OV数据集，每份覆盖度特征均通过统计每个10kb区间内，比对起始位点位于该区间的测序读段数得到。随后通过大小为100个区间的滑动平均对特征进行平滑处理，并记录每个区间位置处滑动窗口的标准差。每个压缩包包含单个域的全部覆盖度特征，每份覆盖度特征以压缩TSV文件存储，其列信息如下： "BINDEX"：区间索引，本质为行号。 "CHR"：区间所在染色体。 "MEAN"：以当前区间为中心的100个连续区间窗口内的平均归一化覆盖度。每个区间大小为10kb。 "SD"：同理计算得到的标准差。 本数据集还包含4个辅助文件： gc-content-1000kb.csv：参考基因组HG38每个区间的GC含量。该文件的行数对应覆盖度特征文件中移除性染色体与线粒体区间后的区间总数，其中-1对应黑名单区域。 mappability-1000kb.csv：每个区间的比对唯一性（mappability）。 blacklisted-10kb-bins.csv：被列入黑名单的10kb区间列表。 D11-D12-batches.json：FRAG数据集的样本名称与文库制备周期（即月份）的映射关系。测序批次严格包含于文库制备周期内，该信息用于留一批次交叉验证，以避免批次效应导致性能评估过于乐观。 GC含量与比对唯一性信息取自：https://github.com/broadinstitute/ichorCNA/tree/master/inst/extdata