| Data processing mimics Buergel et. al, Nat Med 2022 (https://www.nature.com/articles/s41591-022-01980-3) | |
| All detail for endpoints, covariates, metabolites can be viewed in | |
| https://docs.google.com/spreadsheets/d/1XqFFPq3HC2j1hydRc2rUG-SaEXtChmdjcXrJ3XU87To/edit?usp=sharing | |
| 1. Endpoints/Outcome | |
| Binary outcome: /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/ICD10_time_to_event.tab (***) | |
| Generated from ICD10 code, including f.eid, date of birth, sex, age at death, baseline visit, censor time, death date, | |
| 24 clinical endpoint status (binary), and corresponding event time. See detail for 24 endpoints in google sheets above. | |
| Survival residual outcome: /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/surv_res_phase2/surv_res_*.tab (***) | |
| survival residuals are obtained by adjusting three different covariates sets: age+sex, ASCVD (8 CVD related predictors) and PANEL (20 comprehensive predictors). Suffix "_overlap" means only include patients with full record of both metabolomics and proteomics (sample size ~22k), suffix "_all" represents the patients with full metabolomics data (sample size ~277k) | |
| 1. Covariates | |
| Original file: /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/final.predictors.tab | |
| imputed by MICE: /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/covariates_imp_new.tab (***) | |
| included 40 covariates (see covariates detail in google sheets). | |
| The imputed version only include individuals with either metabolites record or proteomics | |
| 2. Metabolites | |
| Original file: /proj/yunligrp/users/altapia/UKB_MWAS/ukb_metabofilter_complete_eur.txt | |
| NatMed including 161 metabolites (see them in google sheets). | |
| We have 159 available metabolites in phase 2 data, sample size ~ 277k: | |
| /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/metabolites_phase2_keep.tab (***) | |
| 3. proteomics | |
| A raw data: /proj/yunligrp/UKBB_phen_29983/olink_data_oct_2023.txt | |
| The QCed data: /proj/yunligrp/users/quansun/UKB_prot/ukbb_olink_data_oct_2023_ins_index_0_std_imp_invn.txt.gz | |
| Note: be sure to include "batch" as one covariate. | |
| /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/proteomics_full.tab (***) | |
| 4. idsplit with different random states | |
| See ID split for 5 folds in folder: /proj/yunligrp/users/djw/UKB_multiomics/rep_NM/surv_res_phase2/idsplit_*.tab (***) | |
| If you are only using individuals with full records in both metabolomics and proteomics, please use "idsplit_overlap", sample size ~22k for most traits (decrease to ~ half in sex specific traits: breast cancer and prostate cancer); | |
| Else if you tend to use only metabolomics data, please use "idsplit_full", sample size ~277k with individuals have full record of metabolomics record. |