OmicsPred as a centralised resource for genetic prediction of multi-omic traits
Carles FoguetiD, Laurent GiliD, Yu XuiD, Sofia Salazar-MagañaiD, Scott C. RitchieiD, Elodie PersyniD, Hae Kyung ImiD, Michael InouyeiD, Samuel A. LambertiD
Abstract
Genetic prediction of multi-omic data has emerged as a cost-effective alternative to direct omics profiling, particularly useful for identifying molecular features associated with disease susceptibility. However, despite its popularity, multi-omic imputation models are fragmented across studies, hindering findability, accessibility, interoperability and re-use. To address this, we developed OmicsPred (https://www.omicspred.org), a centralised platform for the deposition and dissemination of genetic prediction models of multi-omic traits. OmicsPred unifies the most commonly used molecular imputation models (e.g. from PredictDB) and other published studies totalling 3,339,469 prediction models spanning transcriptomic, proteomic, and metabolomic traits (as of May 2026). Each model is accompanied by metadata describing score development and predictive performance, and distributed in formats compatible with popular analytic tools, such as PGS Catalog Calculator and MetaXcan. To demonstrate the utility of the resource for systematic target discovery, we perform a multi-omic phenome-wide association analysis in Million Veterans Program data.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 2
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 2 (11 Jun 2026) · CC BY 4.0
OmicsPred collects 3,339,469 genetic prediction models for transcript, protein and metabolite traits with harmonised metadata and MetaXcan/PGSC-Calculator files, plus multi-ancestry Million Veterans Program phenome-wide association results.
Genotype-based predictors of omic traits let researchers impute molecular levels in biobanks or run transcriptome- and proteome-wide association studies (TWAS, PWAS) on GWAS summary statistics, yet the models are scattered across supplements and study websites with non-standard files and metadata. The GWAS and Polygenic Score Catalogs do not cover molecular-trait models or link to TWAS, PWAS and metabolome-wide (MWAS) frameworks, which is the gap this repository targets.
Key results
ShownAs of May 2026 the resource holds 3,339,469 genetic prediction models for transcriptomic, proteomic and metabolomic traits. Most are PredictDB GTEx v8 models covering gene expression and splicing in 49 human tissues.
The bulk of the count comes from the GTExV8 Elastic Net and MASHR PredictDB datasets. Each score lists its training cohort, sample size, genetic ancestry and R^2 in held-out or external data. Predicted entities are mapped to Ensembl, UniProt or ChEBI, and tissues to EFO. · Abstract; The OmicsPred resource
ShownBeyond GTEx, the resource adds blood and plasma models across several omic platforms and ancestries. These include UK Biobank Olink protein scores trained in a European-ancestry subset and in a multi-ancestry subset.
17,227 INTERVAL models (blood transcripts; Olink and SomaLogic proteins; Nightingale NMR and Metabolon MS metabolites). 3,578 TOPMed and 2,733 ARIC SomaLogic plasma protein models. UK Biobank Olink: 2,612 models trained in 34,557 European-ancestry participants and 2,704 in a multi-ancestry subset of 52,315. · The OmicsPred resource
ShownRunning OmicsPred blood and plasma models through MetaXcan on Million Veterans Program GWAS summary statistics gave more than 190,000 significant associations between imputed transcripts or proteins and disease outcomes, across three ancestry groups.
Up to 1,233 PheCodes in African, Admixed American and European ancestry groups; >190,000 associations at FDR<0.05 out of ~46M tests (Table S1). The results are deposited in OmicsPred next to a previously published UK Biobank PheWAS that used PGSC-Calculator imputed scores. · Phenome-wide insights using OmicsPred
SuggestedMapping scores across datasets recovered known gene-disease links across omic layers and ancestries. It also flagged AGER with hypothyroidism as one of the most consistent signals.
Known links that came back include PCSK9 for coronary artery disease, APOE for hyperlipidemia, and CFH for age-related macular degeneration (Figure 2). AGER with hypothyroidism was significant in 7 of the 9 analysed datasets. The authors cite earlier experimental reports of dysregulated blood AGER in hypothyroidism but run no new tests. · Phenome-wide insights using OmicsPred
SuggestedReactome mappings placed AGER in an NF-κB activation pathway. Other scores in that pathway (NFKB2, NFKBIA, APP) were also associated with hypothyroidism.
AGER maps to Reactome R-HSA-445989, where it acts as a receptor that triggers NF-κB complex activation when a ligand binds. Scores for NFKB2, NFKBIA and APP (an AGER ligand) in the same pathway were significant for hypothyroidism. The main text gives no effect sizes or pathway-level statistics. · Phenome-wide insights using OmicsPred
Methods, data and reproducibility
A curated repository harmonises genotype-based prediction models for transcripts, proteins and metabolites and annotates them (training cohort, ancestry, R^2, Ensembl/UniProt/ChEBI, EFO, Reactome). It serves them by web search, PGSC-Calculator/MetaXcan downloads and REST API, alongside a PheWAS catalogue; MetaXcan TWAS/PWAS on Million Veterans Program GWAS summary statistics demonstrates its use.
- GTEx v8 PredictDB Elastic Net and MASHR models for gene expression and splicing · human (GTEx; text says 49 human tissues) · n = 49 tissues
- INTERVAL cohort models for blood transcripts, Olink/SomaLogic proteins and Nightingale/Metabolon metabolites · human (inferred from cohort name and blood traits) · n = 17,227 models
- TOPMed Multi-omics pilot SomaLogic plasma protein models · human (TOPMed; diverse genetic ancestries) · n = 3,578 models
- ARIC study SomaLogic plasma protein models · human (ARIC cohort) · n = 2,733 models
- UK Biobank Olink proteomics models, European-ancestry subset · human (UK Biobank participants) · n = 2,612 models; 34,557 participants
- UK Biobank Olink proteomics models, multi-ancestry subset · human (UK Biobank participants) · n = 2,704 models; 52,315 participants
- Million Veterans Program harmonised GWAS summary statistics for PheCode outcomes · human (text: human phenome) · n = up to 1,233 PheCodes in AFR, AMR and EUR groups
- Previously published UK Biobank PheWAS of PGSC-Calculator imputed molecular scores · human (UK Biobank)
- Methods & tools
- PredictDB, MetaXcan / S-PrediXcan, PGS Catalog Calculator (PGSC-Calculator), Ensembl, UniProt, ChEBI, Experimental Factor Ontology (EFO), Mondo Disease Ontology, Reactome, GWAS Catalog, Olink and SomaLogic proteomics platforms, Nightingale NMR and Metabolon metabolomics platforms
- Data & accessions
- https://www.omicspred.org/
- Reproducibility gaps
- No source code repository is cited for the platform, the score harmonisation or the MVP TWAS/PWAS analysis. The text does not say how the MVP summary statistics were accessed. Table S1 and Supplemental Notes 1 to 3 are referenced but are not part of the provided text.
Limitations (3)
- Most hosted scores were built in cohorts of predominantly European ancestry. The authors say they will prioritise models from global populations as these become available. · stated by the authors
- The PheWAS demonstration relies on summary statistics from one cohort (MVP) with FDR control only. Neither the AGER-hypothyroidism signal nor its NF-κB pathway context is replicated externally or tested experimentally. · noted in this summary
- Associations are tested with whole scores. Cis-partitioning appears only as a planned feature, so the catalogue cannot yet separate cis-acting from trans or pleiotropic contributions when interpreting associations causally. · noted in this summary
Who should read:Read this if you run TWAS, PWAS or metabolite-level association tests and want harmonised, ancestry-annotated omic prediction weights ready for MetaXcan or PGSC-Calculator, or a searchable catalogue of molecular PheWAS hits.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- An atlas of genetic scores to predict multi-omic traits
Yu Xu, Scott C. Ritchie, … Michael Inouye·NaturePeer-reviewed·Mar 29, 2023
- Improving population-scale disease prediction through multi-omics integration
Esther S. Ng, Ruth Nanjala, … Yang Luo·medRxiv·Nov 24, 2025
- Cross-Platform Omics Prediction procedure enables precision medicine in patients with stage-III melanoma
Kevin Y. X. Wang, Gulietta M. Pupo, … Jean Yee Hwa Yang·npj Digital MedicinePeer-reviewed·Jul 4, 2022
- Multi-omics prediction of oat agronomic and seed nutritional traits across environments and in distantly related populations
Haixiao Hu, Malachy T. Campbell, … Jean‐Luc Jannink·Theoretical and Applied GeneticsPeer-reviewed·Oct 13, 2021
- Integrative machine learning approaches for predicting disease risk using multi-omics data from the UK Biobank
Oscar Aguilar, Cheng Chang, … Manuel A. Rivas·bioRxiv·Apr 16, 2024
- Phenotype prediction using biologically interpretable neural networks on multi-cohort multi-omics data
Arno van Hilten, Jeroen van Rooij, … Gennady V. Roshchupkin·npj Systems Biology and ApplicationsPeer-reviewed·Aug 2, 2024