What do five type 2 diabetes genetic signals point to?

Project abstract

Genome-wide association studies can identify regions associated with type 2 diabetes, but a regional signal does not by itself identify the responsible variant or gene. This project organises existing public evidence from Open Targets release 26.09 for five selected regions from the FinnGen R12 type 2 diabetes study. It brings together 65 fine-mapped variant rows, 11 Locus-to-Gene candidates and 44 molecular-QTL colocalisation rows while preserving source identifiers, missing values, pagination checks, response hashes and quality-control warnings.

Variant PIP, L2G score, COLOC-PIP H4 and eCAVIAR CLPP remain separate because they describe different models and cannot be treated as one causal probability. The result is an auditable explorer and downloadable dataset for examining how variant, gene-ranking and molecular evidence differ across regions. It is a bounded descriptive case study, not causal-gene validation, a clinical target ranking or an analysis of participant-level data.

Read the full problem, workflow, importance and limitations · Start with the beginner and interview guide

Is this regression or classification?

The explorer itself is neither. It is a post-GWAS evidence-integration and audit project. It does not train a model or predict whether a person has diabetes. It retrieves public aggregate results, verifies them, matches related records and presents the evidence for comparison.

Read the detailed beginner guide with interview answers and claims to avoid

How to read this project

The page follows the same question through four steps. Each step narrows or compares evidence; none of the steps proves that a gene causes diabetes.

1. Start with a GWAS signalA GWAS identifies a genomic region associated with type 2 diabetes. The association is a starting point, not a biological explanation.
2. Narrow the variantsFine-mapping assigns source probabilities to variants in the region and returns a credible set of plausible variants.
3. Rank candidate genesOpen Targets L2G uses existing genetic and functional features to rank returned genes on a 0–1 model score.
4. Compare molecular signalsColocalisation asks whether the disease signal and a molecular-QTL signal may involve the same underlying variant.

Evidence availability differs across the five regions

The chart counts records returned by the source. A larger count means that more records were available to inspect; it does not mean stronger biological support.

How to interpret these results

• L2G, PIP, H4 and CLPP answer different questions. They are displayed side by side and are not combined into a new probability.
• A record that was not returned stays missing. It is not changed to zero and is not evidence against a gene.
• Repeated tissues and studies are not independent confirmations. A maximum value is descriptive over the fetched rows.
• The returned candidate list is filtered by Open Targets. “Nearest returned candidate” does not mean the nearest gene across the genome.
• The five regions are a bounded convenience sample from one Finnish study. They do not measure model accuracy or generalise across cohorts, ancestries or diseases.
• Every selected region carries the unresolved source warning “Study has quality control flag(s).” The cached response does not explain the flag, so this project does not guess its effect.

Beginner glossary

Think of the genome as a library: chromosomes are volumes, DNA bases are letters, a gene is an instruction, a variant is a spelling difference and a locus is a marked neighbourhood. These definitions describe how the terms are used on this page.

DNA

The molecule that stores biological instructions using four bases represented as A, C, G and T.

Genome

The complete set of DNA instructions in a person or organism. This project uses public study summaries, not anyone's complete genome.

Chromosome

One long DNA molecule packaged with proteins. Chromosome numbers and positions provide genomic addresses.

Gene

A DNA region used to make a functional RNA or protein. The nearest gene to a signal is not automatically responsible for the disease.

Gene expression

The process of using a gene's information to produce RNA. Expression measurements describe how much a gene is being read in a tissue or cell context.

Phenotype

A measured trait or condition. Type 2 diabetes case/control status is binary; gene-expression level is quantitative.

GWAS

A study that tests many genetic variants across many people to find statistical associations with a trait or disease. Association does not establish causation.

Variant

A DNA sequence difference. A SNP is a one-base variant. An associated variant is not automatically the biological cause.

Linkage disequilibrium

Correlation between nearby variants that tend to be inherited together. It is one reason a GWAS often identifies a region rather than one certain variant.

Locus

A region of the genome around an association signal.

P-value

How incompatible the observed result is with the model's no-association assumption. It is not the probability that a variant is causal.

Lead variant

The representative variant reported for the source signal. It is not automatically causal.

Fine-mapping

Statistical analysis that narrows an associated region to a smaller set of plausible variants.

Credible set

A group of plausible variants constructed to contain the causal variant with a stated probability, commonly 95%, under the fine-mapping model and assumptions.

PIP

Posterior inclusion probability: the source fine-mapping model's support for a variant within that signal. It is not a person's disease risk or a gene-causality probability.

L2G score

Open Targets' 0–1 machine-learning score for ranking candidate genes at a GWAS signal. This project displays the score; it did not train the model.

QTL

Quantitative trait locus: a genomic location where genetic variation is associated with a numerical measurement such as expression, splicing or protein abundance.

eQTL / pQTL / sQTL

QTLs for gene-expression level, protein abundance and RNA splicing respectively. Their effects may depend on tissue or cell type.

Colocalisation

A statistical comparison asking whether a disease association and a molecular-QTL association may share the same underlying variant. It does not prove a causal gene.

H3

For the source COLOC-PIP method, support for both traits having signals in the region but through different variants.

H4

For the source COLOC-PIP method, support for the model in which both traits associate through one shared variant.

CLPP

The eCAVIAR colocalisation statistic. It is method-specific and is not interchangeable with H4.

TSS

Transcription start site: the genomic position where transcription of a gene begins.

Ensembl ID

A stable database identifier for a gene. The project uses it to match candidate genes to molecular-QTL records more reliably than gene names alone.

API and hash

An API supplies structured source data. A SHA-256 hash is a digital fingerprint used here to detect changes to a saved response.

Not returned

The saved source response did not contain a corresponding record or value. It does not mean zero.

Open the expanded glossary and interview guide

Methods and provenance

This project retrieved existing source results, checked response integrity and completeness, preserved missing values, derived lead-variant-to-TSS distances, and built the downloadable tables and explorer. It did not run a new GWAS, fine-mapping model, expression analysis or causal-gene model.