Genomic Prediction Tool

Single variant or batch VCF file prediction

Select the reference genome version

🔑 AlphaGenome API key (optional — needed only for variants not already cached)

Variants already in the MERGE precomputed cache are scored with AlphaGenome without a key. A variant that is not cached needs a live call to Google DeepMind's API, and that call is made under your key — we never spend a MERGE-owned key on your request. Without a key those variants simply come back with the AlphaGenome features marked unavailable; every other model still runs. Get a free key at deepmind.google.com/science/alphagenome.

The AlphaGenome API is provided for non-commercial use only and is subject to the AlphaGenome Terms of Service. Outputs generated by AlphaGenome should not be used for the training of other machine learning models.

Predictions are for theoretical modelling and research purposes only; they should not be used for clinical decision-making or relied upon for medical or other professional advice.

Select Models to Run:

Example: chr17, chr3, 1

Example: 7577121, 123456

Example: A, G

Example: T, C

Format: Chr-Position-Reference-Alternate (e.g., 1-66565-A-T or chr1-66565-A-T)

Prediction Results

Variant:
⚠️ Multiple Transcripts Detected:

🧬 MERGE Pathogenicity Prediction

MERGE is the ensemble pathogenicity probability from the coding, splicing, non-coding, or non-SNV model. Its panels contain no AlphaGenome feature because AlphaGenome's official terms do not permit its outputs to be used as training features for machine-learning models.
MERGE-final combines MERGE with a calibrated single-feature AlphaGenome score (0.5 × MERGE + 0.5 × AlphaGenome) and uses its own interpretation bands. If AlphaGenome is unavailable, MERGE-final is reported as unavailable.
AlphaGenome is still fully extracted and displayed below; it feeds MERGE-final only.
Note on two features (RTX 3090 hardware limits): evo2_1b_base is not among the computed features on this deployment, so it is median-imputed from the training distribution. carbon uses the same SUM log-likelihood-delta scoring as training, but over an 8 192 bp context window instead of the training's 24 576 bp (the larger window exceeds the 24 GB GPU memory). These two features are therefore approximate; all other features are computed exactly.

AlphaGenome Results

HyenaDNA Results

NT Model Results

AlphaMissense Results

ESM-1b Results

GPN-MSA Results

popEVE Results

Evo2 (Local) Results

Evo2 queries the local model with the upstream genomic context to obtain scores. LLR = Reference Score - Variant Score.

Enformer Results

GENERATOR Results

GENERATOR-v2 Results

NT-v2 Results

Carbon Results

Carbon-3B, served by the local GPU gateway (SNV + non-SNV).

✓ Copied to clipboard!

Batch Prediction Results

0
Total Variants Processed
0
Total Transcripts Rendered
Variant (Chr:Pos Ref>Alt) Transcript Gene 🧬 MERGE Score MERGE-final AlphaGenome HyenaDNA NT Score AlphaMissense ESM-1b GPN-MSA popEVE Evo2 Enformer GENERATOR GENERATOR-v2 NT-v2
VariantLiftOver Note Raw MinRaw MeanRaw Max Quantile MinQuantile MeanQuantile Max Splicing Score
VariantLLR ScoreInterpretation
VariantScore (L2)Interpretation
VariantTranscript IDGeneScorePrediction
VariantTranscript IDGeneScorePrediction
VariantLiftOver NoteScore
VariantLiftOver NoteScore
Variant LLR Score Ref Score Var Score Interpretation Context (bp)
VariantScoreInterpretationScore Type
VariantScoreInterpretationScore Type
VariantScore (L2)InterpretationScore Type
Variant LiftOver Note AbsMean Mean Std Max Abs Top5 AbsMean Pos Sum Neg Sum
✓ Copied to clipboard!

Processing Data...

🧬 MERGE — Variant Analysis