← Về thư mục

AlphaGenome Atlas: data model, scorers, AVI, and access

Verified against alphagenome 0.9.0 (released 2026-09-08, the first version with alphagenome.atlas), the Atlas technical report (Cheng, Taylor, Nicolaisen, Pan, Bycroft, Perino, Ward, et al., "AlphaGenome Atlas: in silico mutagenesis of the entire human genome improves prioritization and interpretation of non-coding variants", 2026 preprint), and DeepMind's own Atlas agent skills in google-deepmind/science-skills. Anything marked unverified could not be checked without an API key.

What is in the Atlas

Item	Value
Genome build	GRCh38 / hg38 only. Gene annotations: GENCODE v46.
Variants	Every possible SNV at every non-N reference base (about 9 billion). Over 100 million observed indels (gnomAD v4.1, UK Biobank, All of Us) were scored for the paper, but the public API is currently limited to SNVs; indel expansion is promised "upon final publication".
Model	The distilled AlphaGenome model, 1 Mb input centred on a 128 bp scoring window (so a variant sits within 64 bp of the centre; the paper shows this does not change scores).
Per-variant content	Raw scores for every track of 19 recommended scorer configurations (about 27,000 scalars per variant, about 15,000 without the _ACTIVE scorers), the AVI score with its quantile, 18 SHAP feature attributions, and motif instances.
Motifs	2,601 motifs (94 main TF labels, 122 zinc-finger labels, 464 composite patterns) and about 253 billion motif instances. Browsable on the portal; bulk download promised.
Size	About 1 PB.

Index key semantics

Each record is keyed by chromosome:position:alt on GRCh38. The REF allele is implied by the reference and is not part of the key, so:

• a variant written with REF and ALT swapped (for example minor-allele-first from a GWAS table) is a lookup miss, not an error;
• non-reference to non-reference substitutions do not exist;
• positions where the reference is N were never scored;
• GRCh37 coordinates return wrong-but-plausible answers. Lift over first.

Check REF against the reference FASTA before trusting any Atlas result (genomic-coordinates has normalize_variant.py and check_contigs.py).

The scorers (Table S1 of the report)

requested_scorers= takes these names. Track counts are for human; the _ACTIVE names are the natural spelling of the "(Active)" rows but were not verified against a live scorer_metadata() call.

Name	Scorer class	Parameters	Biosamples	Tracks	Signed	Obs axis
ATAC	CenterMaskScorer	width 501, DIFF_LOG2_SUM	167	167	yes	variant
DNASE	CenterMaskScorer	width 501, DIFF_LOG2_SUM	305	305	yes	variant
CHIP_TF	CenterMaskScorer	width 501, DIFF_LOG2_SUM	163	1,617	yes	variant
CHIP_HISTONE	CenterMaskScorer	width 2001, DIFF_LOG2_SUM	219	1,116	yes	variant
CAGE	CenterMaskScorer	width 501, DIFF_LOG2_SUM	264	546	yes	variant
PROCAP	CenterMaskScorer	width 501, DIFF_LOG2_SUM	6	12	yes	variant
RNA_SEQ	GeneMaskLFCScorer	log fold change over exons	285	371	yes	variant x gene
POLYADENYLATION	PolyadenylationScorer	max log-fold change of isoform ratio	285	371	no	variant x gene
SPLICE_SITES	GeneMaskSplicingScorer	SPLICE_SITES, width None	-	2 (donor, acceptor)	no	variant x gene
SPLICE_SITE_USAGE	GeneMaskSplicingScorer	SPLICE_SITE_USAGE, width None	282	367	no	variant x gene
SPLICE_JUNCTIONS	SpliceJunctionScorer		282	367	no	variant x junction
CONTACT_MAPS	ContactMapScorer		12	28	no	variant
*_ACTIVE (ATAC, DNASE, CHIP_TF, CHIP_HISTONE, CAGE, PROCAP, RNA_SEQ)	same widths, ACTIVE_SUM / GeneMaskActiveScorer	absolute activity of the stronger allele, not a difference	as above	as above	no	as above
AVI_SCORE	composite	one score per variant	-	1	no	variant
AVI_SCORE_FEATURE_IMPORTANCE	SHAP	18 attribution values	-	18	yes	variant

"Signed" scorers can go negative (ALT lowers the signal). For unsigned scorers the direction is not meaningful; only magnitude ranks.

The whole catalogue is 9,440 tracks. python scripts/atlas_query.py tracks --scorer RNA_SEQ --query liver shows the metadata for any of them.

AlphaGenome Variant Impact (AVI)

AVI is a small neural network trained on proxy labels: gnomAD v4.1 variants with a group-maximum filtering allele frequency above 0.001 are "proxy benign" (about 2e7), below 0.001 "proxy impactful" (about 1e8). It therefore learns to rank variants by how strongly negative selection appears to act on them, using 18 features:

Group	Features (var['name'] keys)
AlphaGenome, max absolute effect across all tracks and genes	MAX_ABS_ATAC, MAX_ABS_DNASE, MAX_ABS_CHIP_TF, MAX_ABS_CHIP_HISTONE, MAX_ABS_CAGE, MAX_ABS_PROCAP, MAX_ABS_RNA_SEQ, MAX_ABS_POLYADENYLATION, MAX_ABS_CONTACT_MAPS
AlphaGenome splicing (sites + usage + junctions/5, each max over genes)	MERGED_SPLICING
Protein	ALPHAMISSENSE (0 when there is no AlphaMissense prediction)
Conservation	CACTUS_241_WAY (Zoonomia phyloP, range -20 to 8.9), PHASTCONS_470_WAY
VEP loss-of-function indicators	PROTEIN_TERMINATION (stop gained or frameshift), START_LOST, STOP_LOST
Indel type	IS_INSERTION, IS_DELETION

The 18 SHAP values (expected-gradients approximation, reference = all features zero) sum to the raw AVI score, so the attribution tells you which modality drives a high score: splicing, a TF-binding change, conservation alone, or a coding consequence via AlphaMissense.

Raw, quantile, Phred

The Atlas returns the raw score in X and a cumulative quantile against all genome-wide SNVs in layers['quantiles']. Convert as DeepMind's tools do:

tail  = 1 - cdf_quantile            # fraction of SNVs scoring at least this high
phred = -10 * log10(tail)           # 10 = top 10 %, 20 = top 1 %, 30 = top 0.1 %, 40 = top 0.01 %
top % = 10 ** (-phred / 10) * 100

Indel Phred scores are placed on the SNV quantile curve. The report's guidance: AVI is applicable to coding and non-coding variants, but use region- or application-aware thresholds and prefer ranking over hard cut-offs. Pathogenic regulatory variants land in lower AVI bins than pathogenic protein-truncating or splice-motif variants, so one genome-wide cut-off under-calls regulatory hits.

Benchmarks and case studies (report, for context only)

State-of-the-art on ClinVar non-coding, saturation genome editing, and TraitGym style benchmarks against CADD, GPN-Star, and AlphaMissense after removing training-set overlap; resolution of an epileptic-encephalopathy case in the GREGoR cohort via a DNM1 variant missed by prior tools; a 22 % increase in detectable rare non-coding associations for circulating proteins in UK Biobank.

Client API (alphagenome.atlas.atlas)

from alphagenome.atlas import atlas
from alphagenome.data import genome

client = atlas.create(api_key, timeout=30)          # gRPC to gdmscience.googleapis.com:443

client.scorer_metadata()   # -> {name: ScorerMetadata(name, is_signed, track_metadata: DataFrame)}

client.query_variant(
    genome.Variant.from_str("chr22:36201698:A>C"),  # 1-based position
    requested_scorers=["AVI_SCORE", "AVI_SCORE_FEATURE_IMPORTANCE", "RNA_SEQ"],
    ontology_terms=None,      # e.g. ["UBERON:0001157", "CL:0000084"]; strings or OntologyTerm
    gene_ids=None,            # Ensembl IDs, gene-centric scorers only
    gene_names=None,          # symbols, gene-centric scorers only
)                              # -> {scorer_name: AnnData}

client.query_variants([...], requested_scorers=[...], progress_bar=True, max_workers=10)
client.query_interval(genome.Interval("chr11", 5225726, 5226575), requested_scorers=[...])

• genome.Interval is 0-based half-open; genome.Variant.position is 1-based. Variant.from_str also accepts gnomAD (22-1024-A-C), GTEx (chr22_1024_A_C_b38), and Open Targets spellings via VariantFormat.
• query_interval splits the window into 32 bp chunks, walks the pagination, and returns every SNV in it (3 per base). A 1 kb window is 3,000 variants and takes a few seconds; DeepMind's tool refuses windows above 1,000 bp by default, and so does scripts/atlas_query.py (--max-window).
• query_variants re-raises the first failed lookup; scripts/atlas_query.py queries one variant at a time so a single miss becomes an error cell.
• Each AnnData: X = raw scores (obs x tracks, float32), obs['variant'] = genome.Variant, plus gene_id, gene_name, strand, junction_Start, junction_End for gene-centric scorers; var = track metadata (name, strand, ontology_curie, biosample_name, biosample_type, assay title, gtex_tissue, transcription_factor, histone_mark where applicable); layers['quantiles'] when the server sent calibrated quantiles.
• AVI_SCORE: X shape (n, 1). AVI_SCORE_FEATURE_IMPORTANCE: X shape (n, 18), var['name'] = the feature keys above.
• An ontology_terms filter is ignored for scorers whose tracks have no ontology metadata (the filter string is built that way), so SPLICE_SITES and the AVI scorers still come back.

Errors

atlas.handle_rpc_error maps gRPC status to Python exceptions:

gRPC status	Python	Typical cause
INVALID_ARGUMENT, NOT_FOUND	ValueError	bad key ("API key not valid"), unknown scorer name, variant not in the Atlas (indel, N base, wrong REF)
UNAUTHENTICATED, PERMISSION_DENIED	PermissionError	key without Atlas access, terms not accepted
DEADLINE_EXCEEDED	TimeoutError	60 s per-call timeout
OUT_OF_RANGE	IndexError	interval past the contig end
RESOURCE_EXHAUSTED, UNAVAILABLE	retried by the channel: 5 attempts, exponential back-off 1 s to 60 s	quota or transient outage

An invalid key surfaces as ValueError, not PermissionError (checked live with a dummy key on 2026-09-13).

Access tiers and terms

From the report's Data Availability section and the AlphaGenome terms:

Data	Access	Route
AVI scores, all hg38 SNVs	permissive licence, commercial use allowed	static download (Tabix) at https://alphagenome.google/downloads
AVI scores	non-commercial	API (AVI_SCORE)
AVI feature attributions	non-commercial	static download (Tabix) and API (AVI_SCORE_FEATURE_IMPORTANCE)
Splicing scores	non-commercial	static download (Tabix) and API
All other raw Atlas track scores	non-commercial	API only
Motif compendium and instances	browse on the portal; bulk download promised	portal
Commercial API access	"coming soon" via Google Cloud Model Garden	-

Model and Atlas outputs are for research and theoretical modelling, may not be used to train other machine-learning models, and are not for diagnostic procedures or medical decision-making. The website is free for non-commercial use. Query rates are demand-dependent and unpublished; the docs say the Atlas "will typically have a larger query rate" than on-demand model predictions.

The portal

https://deepmind.google.com/science/alphagenome/atlas (also reachable as https://alphagenome.google/atlas). Query parameters, from DeepMind's link builder:

Parameter	Meaning
q	variant chr:pos:ref>alt (1-based), interval chr:start-end (1-based closed), gene symbol, or Ensembl gene ID. rsIDs are not accepted.
m	view: variant, locus, entity (gene), motifs
i	viewport interval, needed before motif instances render
f	filters: BIOSAMPLE_NAME:K562, BIOSAMPLE_TYPE:... (AND), SCORER_MODALITY:RNA-seq, ASSAY_TRANSCRIPTOR_FACTOR:GATA1, ASSAY_HISTONE_MARK:H3K27ac (OR within the assay group), GENE_NAME:HBB
lItems	layout: avi, section:RNA_SEQ, section:DNASE, section:CHIP_TF, ..., pinned:<track key>
scores, md, tpRenames, tpLegendTitle	the /atlas/track-predictions REF-vs-ALT comparison page

Because RNA-seq and DNase tracks carry no TF code, a filter made only of TF predicates hides them; add SCORER_MODALITY:RNA-seq,SCORER_MODALITY:DNase alongside (scripts/atlas_link.py does this automatically).

Bulk files and annotations

• GENCODE v46 GTF as Feather, the annotation behind both Atlas and model gene scores: https://storage.googleapis.com/alphagenome/reference/gencode/hg38/gencode.v46.annotation.gtf.gz.feather (about 318 MB, roughly 4 GB RAM in pandas; column names are capitalised: Feature, Start, End, Strand). Load with pd.read_feather and use alphagenome.data.gene_annotation helpers (filter_to_mane_select_transcript, get_gene_interval, extract_tss) so gene models match the scores.
• AVI Tabix downloads: https://alphagenome.google/downloads (not inspected here).

Citing

Cite both the Atlas report (above) and the model paper: Avsec Ž. et al., "Advancing regulatory variant effect prediction with AlphaGenome", Nature 649, 1206-1218 (2026), doi:10.1038/s41586-025-10014-0. Software: https://github.com/google-deepmind/alphagenome (Apache-2.0 client; the model weights and outputs carry separate terms).