AlphaGenome 2

Client, endpoints, outputs, conditions and scorers for AlphaGenome 2.

DNA client

models.v2.dna_client.create(api_key, *[, ...])	Creates a model client for a given API key.
models.v2.dna_client.ModelVersion(value)	Enumeration of available model checkpoints.
models.v2.dna_client.Organism(value)	Enumeration of all the available organisms.
models.v2.dna_client.validate_sequence_length(length)	Validate that the input sequence length is supported by the model.
models.v2.dna_client.DnaClient(*, channel[, ...])	Client for interacting with a DNA model server.

Sequence lengths

Supported input sequence lengths, for use with sequence_length arguments and validate_sequence_length(). Input gradient endpoints support only SEQUENCE_LENGTH_100KB and SEQUENCE_LENGTH_1MB.

alphagenome.models.v2.dna_client.SEQUENCE_LENGTH_16KB = 16384

16,384 bp (2^14).

alphagenome.models.v2.dna_client.SEQUENCE_LENGTH_100KB = 131072

131,072 bp (2^17).

alphagenome.models.v2.dna_client.SEQUENCE_LENGTH_500KB = 524288

524,288 bp (2^19).

alphagenome.models.v2.dna_client.SEQUENCE_LENGTH_1MB = 1048576

1,048,576 bp (2^20).

Limits

alphagenome.models.v2.dna_client.MAX_CONDITIONING_CELLS = 256

Maximum number of cells accepted per cell_conditioning call (256).

Endpoints

The DnaClient exposes the following endpoints:

Method	Description
predict_sequence()	Predict tracks for a raw DNA sequence (predict_sequences for batches).
predict_interval()	Predict tracks for a reference genome interval (predict_intervals for batches).
predict_variant()	Predict tracks for the REF and ALT alleles of a variant (predict_variants for batches).
score_variant()	Score the effect of a variant (score_variants for batches).
score_interval()	Score an interval (score_intervals for batches).
score_ism_variants()	Score all single-base variants in an interval (in silico mutagenesis).
predict_interval_gradients()	Compute input gradients of a gene’s predicted expression (RNA_SEQ, SC_GENE_EXPRESSION, or SC_GENE_EXPRESSION_DENOISED) over an interval centered on the gene’s TSS (predict_interval_gradients_batch for batches).
encode_cells()	Encode raw single-cell UMI counts (human) into per-cell embeddings with the cell encoder, without running the sequence model.
supported_conditions()	List reference conditions with the imputation mode (observed, imputed_supported, or imputed_exploratory) of each output type.
observed_metadata()	Get the metadata for tracks observed during training.

DNA output

models.v2.dna_output.OutputType(value)	Enumeration of all the available types of outputs.
models.v2.dna_output.Output([atac, cage, ...])	Model outputs for a single prediction.
models.v2.dna_output.OutputMetadata([atac, ...])	Metadata detailing the content of model output.
models.v2.dna_output.VariantOutput(...)	Model outputs for a variant prediction.
models.v2.dna_output.TargetGeneMetadata(...)	Metadata for a single target gene.
models.v2.dna_output.InputGradientsOutput(...)	Model outputs for an input sequence gradient prediction.

Conditions

Reference conditions (condition=) are built with model_condition; expression conditioning inputs (cell_conditioning=) are prepared with gene_counts_condition.

models.v2.model_condition.ModelCondition([...])	Metadata describing the type of a model condition.
models.v2.model_condition.from_curies(...)	Get a ModelCondition from a list of ontology curies.
models.v2.gene_counts_condition.from_anndata(adata)	Converts AnnData rows to a CellConditioningInput proto.
models.v2.gene_counts_condition.subsample(...)	Subsamples an AnnData object into a mapping of condition label to AnnData.

Variant scorers

models.v2.variant_scorers.AggregationType(value)	Enum indicating the type of variant scorer aggregation.
models.v2.variant_scorers.BaseVariantScorer(value)	Enum indicating the type of variant scoring.
models.v2.variant_scorers.CenterMaskScorer(...)	Variant scorer for center mask scorers.
models.v2.variant_scorers.ContactMapScorer()	Variant scorer for contact map scorers.
models.v2.variant_scorers.GeneMaskLFCScorer(...)	Variant scorer for gene-mask log fold change scoring.
models.v2.variant_scorers.GeneMaskActiveScorer(...)	Variant scorer for gene-mask active allele scoring.
models.v2.variant_scorers.GeneMaskSplicingScorer(...)	Variant scorer for gene-mask scoring for splicing.
models.v2.variant_scorers.MethylationScorer(width)	Variant scorer for CpG methylation.
models.v2.variant_scorers.PolyadenylationScorer()	Variant scorer for polyadenylation QTLs (paQTLs).
models.v2.variant_scorers.SpliceJunctionScorer()	Variant scorer for splice junction scoring.
models.v2.variant_scorers.get_recommended_scorers(...)	Returns the recommended variant scorers for a given organism.
models.v2.variant_scorers.tidy_anndata(adata)	Formats an AnnData score as a tidy DataFrame.
models.v2.variant_scorers.tidy_scores(scores)	Formats scores into a tidy (long) pandas DataFrame.

alphagenome.models.v2.variant_scorers.RECOMMENDED_VARIANT_SCORERS

Immutable mapping from a short name (e.g. 'RNA_SEQ', 'ATAC', 'SC_GENE_EXPRESSION_DENOISED', 'SC_GENE_EXPRESSION') to a variant scorer configured with our recommended settings for that output type. Each output type has a log-fold-change style scorer and, where applicable, an _ACTIVE variant that scores absolute activity. Use get_recommended_scorers() to get the subset supported for a given organism and request type.

alphagenome.models.v2.variant_scorers.VariantScorerTypes: TypeAlias

Union type alias of supported AlphaGenome 2 variant scorer configurations.

Interval scorers

models.v2.interval_scorers.IntervalAggregationType(value)	Enum indicating the type of interval scorer aggregation.
models.v2.interval_scorers.BaseIntervalScorer(value)	Enum indicating the type of interval scoring.
models.v2.interval_scorers.GeneMaskScorer(...)	Interval scorer for gene-mask scoring.

alphagenome.models.v2.interval_scorers.IntervalScorerTypes

Union type alias of supported AlphaGenome 2 interval scorer configurations.