1,500 simulated transcriptomic variants for MINTIE paper

Contains RNA-seq data set of 1,500 simulated heterozygous transcriptomic variants (500 fusions, 500 splice variants and 500 transcribed structural variants) used in the MINTIE paper. An additional 100 unmodified background genes were also added. The controls set contains unmodified sequences of all variant genes included in the case sample. Variant information and paired end reads, as well as the fasta files from which they were generated, are provided. Code used to generate these samples can be found under https://github.com/Oshlack/MINTIE/tree/master/simu. Simulations were generated by extracting sequence from the transcripts listed in the hg38 UCSC RefSeq reference, and simulating reads from the resulting sequence. 100 variants from 15 variant types were generated (five fusion types: canonical, extended exon, novel exon, with insertion and unpartnered, five TSV types: insertions, deletions, ITDs, PTDs and inversions, and five novel splice variants: extended exons, novel exons, truncated exons, skipped exons and retained introns. Only transcripts from genes that did not overlap any other genes were used in the simulation. Additionally, each transcript had to have at least 3 exons to be considered as a simulation transcript. All fusions were simulated by selecting the first two and the last two exons from two random transcripts from different genes, and inserting the intervening sequence. Canonical fusions contained no intervening sequence, while fusions with extended exons inserted 30-200bp of intronic sequence from the end of the second exon of the first transcript. Similarly, fusions with novel exons contained intronic sequence 30-200bp downstream with a size of 30-200bp. Non-canonical fusions with insertions were generated by inserting 7-50bp of randomly-generated sequence between the two fusion transcripts. Small TSVs were generated by inserting, duplicating or deleting sequence within randomly selected exons from randomly selected transcripts. These small variant types were between 7 and 50 base-pairs and had to reside at least 10bp within the exon. Inversions and partial-tandem duplications were generated by selecting 1-3 random exons within a transcript and either inverting or duplicating their sequence in tandem. Lastly, splice variants were generated by extending or placing novel exons downstream of a randomly selected exon. To ensure that novel or extended exons did not overlap exons from other transcripts (or downstream exons of the same transcript), each candidate exon was checked for these potential overlaps (which would otherwise result in obfuscation of the variant, or the wrong variant type being created). Novel junction variants were created by selecting a random pair of exons and checking whether an existing junction existed between them (creating a transcript with this junction if not). Two randomly-selected neighbouring exons were both truncated at their facing ends (end and start respectively) by 30-200bp. Retained introns included a random intronic sequence from a given transcript that was >30bp. The presence of correct splicing motifs was not considered for the simulation. In addition to each variant gene, the sequence to the unaltered wild-type gene was added to the simulated case sample’s reference. An additional 100 unaltered background genes were also added to the case sample. A control sample reference was also generated, which included the unaltered wildtype sequence only for all simulated transcripts. ART-illumina (doi: 10.1093/bioinformatics/btr708) v2.5.8 was run on the corresponding references with 100bp paired-end reads with a fragment size of 300 and coverage of 50 (transcripts should thus have an effective coverage of 100, given the bi-allelic reference containing variant and wildtype transcripts). We also include three down-sampled versions of the simulation files (40x, 20x and 10x) used in the MINTIE paper. Note that the variant coverage will be half the sequence coverage. These were down-sampled using seqtk v1.0 (https://github.com/lh3/seqtk).本数据集包含用于MINTIE研究论文的1500个模拟杂合转录组变异的RNA测序（RNA-seq）数据集，其中涵盖500个融合变异、500个剪接变异及500个转录结构变异。此外还额外加入了100个未修饰的背景基因。对照组包含病例样本中所有变异基因的未修饰序列。本数据集提供变异信息、双端测序读段（paired-end reads），以及用于生成这些数据的FASTA文件。用于生成该样本的代码可访问https://github.com/Oshlack/MINTIE/tree/master/simu获取。模拟流程为：从hg38 UCSC RefSeq参考基因组注释的转录本中提取序列，再基于所得序列模拟测序读段。 15种变异类型共计100个变异被生成：包括5种融合变异类型（经典型、延伸外显子型、新外显子型、插入型及非配对型）、5种TSV变异类型（插入、缺失、内部串联重复（ITD）、部分串联重复（PTD）及倒位），以及5种新型剪接变异类型（延伸外显子、新外显子、截短外显子、外显子跳跃及内含子保留）。模拟仅使用不与其他基因重叠的基因对应的转录本，且每个转录本需至少包含3个外显子方可纳入模拟流程。 所有融合变异的模拟流程为：从两个不同基因的随机转录本中分别选取前两个外显子与后两个外显子，并插入中间序列。经典型融合变异不含中间序列；延伸外显子型融合变异会在第一个转录本第二外显子末端插入30~200bp的内含子序列；新外显子型融合变异则会在下游插入长度为30~200bp的内含子序列；带有插入序列的非经典型融合变异则会在两个融合转录本之间插入7~50bp的随机生成序列。 小型TSV变异的模拟方式为：在随机选取的转录本的随机外显子内部插入、复制或删除序列。这类小型变异的长度介于7~50bp之间，且需至少距离外显子边界10bp。倒位与部分串联重复变异则通过选取转录本内1~3个随机外显子，对其序列进行倒位或串联复制生成。最后，剪接变异的模拟通过在随机选取的外显子下游延伸或插入新外显子实现。为确保新外显子或延伸外显子不会与其他转录本的外显子（或同一转录本的下游外显子）发生重叠，需对每个候选外显子进行潜在重叠检测——否则会导致变异混淆或错误生成变异类型。 新型剪接连接变异的生成方式为：选取一对随机外显子，检测二者之间是否已存在天然剪接连接——若不存在则生成带有该连接的转录本。将两个随机选取的相邻外显子在相对末端（分别为末端与起始端）截短30~200bp。内含子保留变异则选取给定转录本中长度大于30bp的随机内含子序列纳入变异。本次模拟未考虑正确剪接基序的存在性。 除每个变异基因外，模拟病例样本的参考序列中还加入了未修饰的野生型基因序列。此外还向病例样本中额外添加了100个未修饰的背景基因。同时生成了对照组样本参考序列，该序列仅包含所有模拟转录本的未修饰野生型序列。 使用ART-illumina（doi: 10.1093/bioinformatics/btr708）v2.5.8对相应参考序列进行测序模拟，生成100bp双端读段，插入片段长度为300bp，测序覆盖度为50——由于双等位基因参考序列同时包含变异转录本与野生型转录本，因此转录本的有效覆盖度应为100。本数据集还提供了MINTIE论文中使用的3种模拟文件的降采样版本（40×、20×及10×）。需注意：变异位点的覆盖度为序列总覆盖度的一半。这些降采样操作通过seqtk v1.0（https://github.com/lh3/seqtk）完成。