bio-causal-genomics-effector-gene-prioritization

What this skill does

## Version Compatibility

Reference examples tested with: MAGMA 1.10+ (ctglab.nl/software/magma), FUMA web platform v1.6+ (fuma.ctglab.nl), Open Targets Genetics API (REST + GraphQL, June 2024 release), PoPS (head of `FinucaneLab/pops`, 2024), cS2G pre-computed scores (alkesgroup.broadinstitute.org/cS2G/, 2022), ABC-Enhancer-Gene-Prediction 0.2.2+, ENCODE-rE2G v1.0+ (2024), DEPICT v1 rel194, INQUISIT (Fachal 2020 supplementary), Python 3.9-3.11, R 4.3+, PLINK 1.9 + PLINK 2.0.

Before using code patterns, verify installed versions match. If versions differ:
- CLI: `magma --help` to confirm gene-window, gene-annot, and gene-set flag names
- Python: `pip show ot-graphql opentargets-genetics`; introspect endpoints at `api.genetics.opentargets.org/graphql`
- R: `packageVersion('coloc')` etc. for upstream evidence integration

If a script throws an error about an argument that has moved (e.g. an Open Targets endpoint renamed during a release) or a model file schema change, introspect the installed tool and adapt rather than retrying. Open Targets Genetics deprecated the standalone Genetics Portal in 2024 in favour of the integrated platform; verify endpoint URLs at the time of use.

# Effector Gene Prioritization

**"Which gene at this GWAS locus is actually the causal mediator?"** -> Integrate fine-mapping, colocalization, chromatin-based enhancer-gene predictions, distance, and gene-similarity priors into a per-locus per-gene confidence score, then require concordance across multiple orthogonal evidence streams before nominating a causal effector. Effector gene prioritization is the bridge between statistical fine-mapping (variant level) and biological hypothesis (gene level); it is the most failure-prone step in GWAS-to-target pipelines because the nearest-gene assumption is wrong roughly 30-50% of the time at well-studied loci.

- CLI (gene-level association): `magma --bfile ref --gene-loc geneloc.txt --pval gwas.tsv ncol=N --out out` -> `magma --gene-results out.genes.raw --set-annot annot.txt --out out`
- Web (integrative): FUMA SNP2GENE at fuma.ctglab.nl (positional + eQTL + Hi-C + chromatin in one workflow)
- API (pre-computed L2G): Open Targets Genetics GraphQL `studyLocus2GeneTable` query (note: Open Targets Genetics was consolidated into the Open Targets Platform in 2024; verify the live endpoint at `api.platform.opentargets.org/api/v4/graphql`)
- Python (similarity prior): `python pops.py --gene_annot gene_annot.txt --features features --magma_prefix magma_out --out out`
- Lookup (combined SNP-to-gene): cS2G pre-computed gene scores at alkesgroup.broadinstitute.org/cS2G/
- CLI (enhancer-gene): ABC pipeline or ENCODE-rE2G (cross-reference atac-seq/enhancer-gene-linking)

V2G is not one method but a portfolio. Open Targets L2G aggregates per-locus per-gene features (distance + coloc + chromatin + V2G) trained on curated gold-standard genes; PoPS adds an orthogonal genome-wide polygenic prior from gene-pathway co-membership; MAGMA provides the lightweight gene-level p-value baseline. Strong effector calls emerge from concordance across these orthogonal signal types, not from any single tool.

## Algorithmic Taxonomy

| Tool | Model | Inputs | Output | Strength | Fails when |
|------|-------|--------|--------|----------|------------|
| Open Targets L2G (Mountjoy 2021 Nat Genet 53:1527) | Gradient-boosting classifier on per-(locus, gene) features (distance, fine-mapping, coloc, chromatin, V2G) trained on curated gold standards | Pre-computed per study; queried via API | Per-(study, locus, gene) L2G score 0-1 | Most validated integrative scorer; built into Open Targets Platform; updated quarterly | Trait must be in OT release; custom traits need re-training; coverage limited to OT-curated GWAS catalog |
| V2G (Ghoussaini 2021 Nat Genet 53:1530) | Open Targets V2G feature aggregator: per-variant eQTL/sQTL/pQTL + chromatin + distance | OT pre-computed | Per-(variant, gene) score | Variant-resolution; complements locus-resolution L2G | Feature weights are fixed; cannot tune per-trait |
| MAGMA (de Leeuw 2015 PLoS Comput Biol 11:e1004219) | SNP-to-gene window aggregation + multiple regression on summary statistics | GWAS sumstats + gene annotation + LD reference (PLINK bfile) | Gene-level Z, p; gene-set p | Mature, fast, lightweight; supports gene-set enrichment in same pass; widely cited | Window choice (0+0 vs 35kb+10kb vs 50kb+50kb) shifts top genes; cannot detect distal regulation outside window |
| FUMA SNP2GENE (Watanabe 2017 Nat Commun 8:1826) | Web platform combining positional + eQTL + Hi-C + chromatin annotation + MAGMA | Sumstats upload to fuma.ctglab.nl | Annotated locus + prioritised gene table | One-click integrative analysis; no local install needed; community standard for GWAS post-hoc | Web-only; no API for high-throughput; pre-baked annotations may lag latest reference releases |
| cS2G (Gazal 2022 Nat Genet 54:827) | Weighted aggregation of 10 SNP-to-gene strategies (Closest TSS, ABC, fine-mapped eQTL, etc.) calibrated on heritability enrichment | Per-SNP lookup | Combined per-SNP score allocated to genes | Heritability-calibrated; pre-computed gene scores for downstream filtering | Aggregation weights are population-averaged; cell-type-specific signal averaged out; coverage limited to baseline-LF SNP universe |
| PoPS (Weeks 2023 Nat Genet 55:1267) | LASSO regression of per-gene MAGMA Z on genome-wide gene-feature matrix (pathway membership, co-expression, PPI) | MAGMA Z + gene-feature matrix | Per-gene priority score (PoPS); per-locus relative ranking | Orthogonal to distance / proximity; identifies genes with similar pathway / co-expression profile to other GWAS hits | Pathway co-membership similarity is similarity-based; can hand-feed bias if features are not curated; complementary to L2G, not redundant |
| FLAMES / Funmap2 (Lake 2024) | Combined per-feature scoring with deep learning + integrative scoring | Sumstats + features | Per-gene prioritisation | Recent integrative method | Limited validation outside the publication test set; method choice still evolving |
| INQUISIT (Fachal 2020 Nat Genet 52:56) | Three-level scoring for coding, regulatory-proximal, regulatory-distal; trait-specific (breast cancer) | Sumstats + cancer-specific annotation panel | Per-gene INQUISIT score | Cancer-tuned; integrates expression and chromatin context | Originally trait-specific (breast cancer); adapting to other diseases requires re-curation |
| DEPICT (Pers 2015 Nat Commun 6:5890) | Empirical Bayes; gene set enrichment + tissue prioritisation + reconstituted gene sets | Sumstats | Per-gene p; pathway enrichment; tissue priority | Old but still cited; combines three useful outputs | Reconstituted gene sets are dated (2015 expression panel); largely superseded by L2G + PoPS combination |
| ABC + ENCODE-rE2G | Activity x Contact enhancer-gene model (Fulco 2019) and logistic-regression refinement (ENCODE 2024) | ATAC + H3K27ac + Hi-C/Micro-C | Per-(enhancer, gene) score | Direct mechanistic enhancer-gene link in matched cell type; gold-standard for distal regulation | Requires matched epigenome data; cell-type-specific; covered in detail in atac-seq/enhancer-gene-linking |
| sc-eQTL + cell-type-specific TWAS (e.g. Yazar 2022 OneK1K) | Per-cell-type eQTL panels + per-cell-type prediction weights | sc-eQTL panel + sumstats | Cell-type-resolved gene candidates | Resolves cell-type-specific causal genes that bulk-tissue TWAS averages out | Requires matched single-cell eQTL panel; not yet pre-built for most cell types |

Methodology evolves; verify against the current Open Targets release (platform-docs.opentargets.org), the latest PoPS feature matrix at FinucaneLab/pops, and ABC / ENCODE-rE2G releases before locking on a single prioritiser. The L2G + PoPS combination is the current de facto two-method baseline; cS2G is the heritability-calibrated lookup; FUMA is the no-install community standard.

## Decision Tree by Scenario

| Scenario | Recommen