AlphaGenome Cardiovascular Non-Coding Variant Analysis — Data Deposit

This deposit contains all intermediate and final data artifacts for the manuscript "Systematic Characterization of Cardiovascular Non-Coding Variants Using AlphaGenome". We used Google DeepMind's AlphaGenome model (Nature, 2026) to score 200,813 cardiovascular non-coding variants from ClinVar (n=15,623) and fine-mapped Open Targets GWAS (n=185,198) across 19 variant scorers at 1 Mb context, single-base resolution. Filtering to cardiac-relevant GTEx tissues and ENCODE biosamples retained 452.7 million cardiac score rows covering 104,473 unique variants with regulatory signal. Multimodal analysis identified a convergent LMNA promoter VUS cluster as a high-priority candidate set for functional validation under ACMG PP3. File organization. Files are grouped by numeric prefix corresponding to pipeline stage: 01 — Input variants (curated catalogues before AlphaGenome scoring): clinvar_cardio_extract.tsv — ClinVar cardiovascular non-coding variants after filtering (15,623 variants) gwas_cardio_credible_sets.tsv — Open Targets fine-mapped GWAS credible-set variants (185,198) merged_cardio_variants.tsv — Deduplicated merge of the two sources (200,813) scoring_intervals.tsv — 1 Mb intervals centred on each variant, used as AlphaGenome input 02 — AlphaGenome scores (variant scoring across all 19 recommended scorers): all_scores.parquet — Full unfiltered scoring output (3,572,509,053 rows) cardiac_scores.parquet — Cardiac-tissue-filtered subset (452,688,226 rows; GTEx heart/artery + ENCODE cardiac biosamples) 03 — Analysis outputs (post-processing feeding directly into manuscript Figures and Tables): variant_multimodal_summary.tsv — Per-variant summary across 6 modality groups (Fig 3) modality_enrichment.tsv — Per-modality strong-signal rates (Fig 3C) multimodal_hits.tsv — Variants with signal in ≥2 modalities (Fig 2A, Fig 3A) ranked_variant_table.tsv — Composite-score-ranked full catalogue (Fig 4, Tables 2 and 4) gwas_direction_of_effect.tsv — Direction-of-effect calls for all variant–gene pairs (Fig 5, Table 3) gwas_direction_of_effect_cardiac.tsv — Filtered to cardiac-trait pairs (Fig 5, Table 3) 04 — Manuscript supplementary tables (Tables S1–S4 as deposited with the manuscript): supp_S1_gene_panel.tsv — Curated cardiovascular gene panel (101 genes, 10 categories) supp_S2_trait_keywords.tsv — Cardiovascular-trait include/exclude keyword lists supp_S3_full_ranked_variants.{tsv,xlsx} — Full ranked variant table supp_S4_doe_cardiac_full.tsv — Cardiac-trait DoE pairs 05 — Vignettes (mechanistic ISM and REF/ALT figures for Figure 6): ism_logo_*.png — ISM sequence-logo plots ism_*.npy — Raw per-base ISM effect arrays (numpy format) ref_alt_*.png — REF vs ALT predicted RNA-seq track plots Vignettes cover the LMNA promoter VUS cluster (chr1:156115264, chr1:156115279, chr1:156115286), DUSP1 (chr5:172770036), and VIM (chr10:17229111). Schema for parquet files (all_scores.parquet, cardiac_scores.parquet): variant_id (string) — chr:pos:ref>alt output_type (string) — AlphaGenome scorer (RNA_SEQ, SPLICE_JUNCTIONS, CHIP_HISTONE, CAGE, DNASE, CHIP_TF, SPLICE_SITE_USAGE, CONTACT_MAPS, ATAC, PROCAP) gene_name (string) — target gene gtex_tissue (string, nullable) — GTEx tissue for RNA-seq/splicing rows biosample_name (string, nullable) — ENCODE biosample for accessibility/TF/histone/contact map rows raw_score, quantile_score (float) — AlphaGenome predictions variant_scorer, track_name, biosample_type — scoring metadata Reproducibility. SHA-256 checksums for every file are in CHECKSUMS.sha256. Verify with sha256sum -c CHECKSUMS.sha256. Companion code: https://github.com/patricio-astudillo/alphagenome-cardio (MIT license; 10.5281/zenodo.19591759). Upstream data sources: ClinVar variant_summary.txt.gz, NCBI release 2024-03-31, retrieved 2026-04-05 from https://ftp.ncbi.nlm.nih.gov/pub/clinvar/tab_delimited/ Open Targets Platform release 26.03, queried via GraphQL at https://api.platform.opentargets.org/api/v4/graphql AlphaGenome model and API: https://github.com/google-deepmind/alphagenome本数据集包含论文《使用AlphaGenome系统表征心血管非编码变异》的所有中间与最终数据产物。 本研究使用谷歌DeepMind开发的AlphaGenome模型（《Nature》，2026），以1 Mb侧翼序列、单碱基分辨率的19种变异评分器，对来自ClinVar（Clinical Variation Database，临床变异数据库）和经精细定位的Open Targets全基因组关联分析（Genome-Wide Association Study，GWAS）的200,813个心血管非编码变异进行评分。经筛选匹配心脏相关的基因型-组织表达（Genotype-Tissue Expression，GTEx）组织及DNA元件百科全书（Encyclopedia of DNA Elements，ENCODE）生物样本后，最终得到4.527亿条心脏评分记录，覆盖104,473个带有调控信号的独特变异。多模态分析鉴定出一个收敛的LMNA启动子意义未明变异（Variant of Uncertain Significance，VUS）簇，该变异簇符合美国医学遗传学与基因组学学会（American College of Medical Genetics and Genomics，ACMG）标准PP3，可作为功能验证的高优先级候选集合。 ### 文件组织结构 文件按对应分析流程阶段的数字前缀分组： 01 — 输入变异（AlphaGenome评分前的精选变异目录）： - `clinvar_cardio_extract.tsv`：经筛选后的ClinVar心血管非编码变异（共15,623个变异） - `gwas_cardio_credible_sets.tsv`：Open Targets精细定位的GWAS可信集变异（共185,198个） - `merged_cardio_variants.tsv`：两个来源去重后的合并变异集（共200,813个） - `scoring_intervals.tsv`：以每个变异为中心的1 Mb侧翼区间，用作AlphaGenome的输入数据 02 — AlphaGenome评分（基于全部19种推荐评分器的变异评分结果）： - `all_scores.parquet`：完整未过滤的评分输出（共3,572,509,053条记录） - `cardiac_scores.parquet`：经心脏组织过滤后的评分子集（共452,688,226条记录，对应GTEx心脏/动脉组织及ENCODE心脏生物样本） 03 — 分析输出（直接用于论文图表与表格的后处理结果）： - `variant_multimodal_summary.tsv`：基于6个模态组的单变异汇总数据（对应图3） - `modality_enrichment.tsv`：各模态强信号占比（对应图3C） - `multimodal_hits.tsv`：在≥2个模态中检测到信号的变异（对应图2A、图3A） - `ranked_variant_table.tsv`：基于综合评分排序的完整变异目录（对应图4、表2与表4） - `gwas_direction_of_effect.tsv`：所有变异-基因对的效应方向判定结果（对应图5、表3） - `gwas_direction_of_effect_cardiac.tsv`：经心脏性状对筛选后的上述效应方向结果（对应图5、表3） 04 — 论文补充表格（随论文提交的表S1~S4）： - `supp_S1_gene_panel.tsv`：精选心血管基因集（共101个基因，分为10个类别） - `supp_S2_trait_keywords.tsv`：心血管性状相关关键词的包含/排除列表 - `supp_S3_full_ranked_variants.{tsv,xlsx}`：完整排序的变异表 - `supp_S4_doe_cardiac_full.tsv`：心脏性状相关的效应方向变异-基因对全集 05 — 示例插图（对应图6的机制性计算机饱和诱变（In Silico Saturation Mutagenesis，ISM）与参考/等位基因图谱）： - `ism_logo_*.png`：ISM序列Logo图 - `ism_*.npy`：单碱基ISM效应数组原始数据（numpy格式） - `ref_alt_*.png`：参考等位基因与替代等位基因的预测RNA-seq轨道图 本示例插图覆盖LMNA启动子VUS簇（chr1:156115264、chr1:156115279、chr1:156115286）、DUSP1（chr5:172770036）及VIM（chr10:17229111）三个区域。 #### Parquet文件（`all_scores.parquet`、`cardiac_scores.parquet`）字段说明： - `variant_id`（字符串型）：变异标识符，格式为chr:pos:ref>alt - `output_type`（字符串型）：AlphaGenome评分器类型，涵盖RNA_SEQ、SPLICE_JUNCTIONS、CHIP_HISTONE、CAGE、DNASE、CHIP_TF、SPLICE_SITE_USAGE、CONTACT_MAPS、ATAC、PROCAP - `gene_name`（字符串型）：靶标基因 - `gtex_tissue`（字符串型，可空）：用于RNA-seq/剪接数据分析的GTEx组织 - `biosample_name`（字符串型，可空）：用于可及性/转录因子/组蛋白/接触图谱数据分析的ENCODE生物样本 - `raw_score`、`quantile_score`（浮点型）：AlphaGenome预测得分 - `variant_scorer`、`track_name`、`biosample_type`：评分元数据 ### 可复现性说明 所有文件的SHA-256校验和均存储于`CHECKSUMS.sha256`文件中，可通过`sha256sum -c CHECKSUMS.sha256`命令完成校验。 ### 配套代码 访问地址：https://github.com/patricio-astudillo/alphagenome-cardio（采用MIT许可证；DOI：10.5281/zenodo.19591759）。 ### 上游数据源 1. ClinVar变异摘要文件`variant_summary.txt.gz`，NCBI 2024-03-31版本，于2026-04-05从https://ftp.ncbi.nlm.nih.gov/pub/clinvar/tab_delimited/ 下载获取 2. Open Targets Platform 26.03版本，通过GraphQL接口https://api.platform.opentargets.org/api/v4/graphql 查询获取 3. AlphaGenome模型与API：https://github.com/google-deepmind/alphagenome