File size: 7,888 Bytes
# Copyright 2026 Google LLC.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Example script for AlphaGenome variant scoring.

This version avoids failing when the model was created without annotation-backed
variant scorers such as GENE_MASK_LFC. It keeps the recommended scorers that
are actually available in the current model instance and skips the rest.
"""

from collections.abc import Sequence
import pathlib
import sys

_PROJECT_ROOT = pathlib.Path(__file__).resolve().parents[1]
if str(_PROJECT_ROOT) not in sys.path:
  sys.path.insert(0, str(_PROJECT_ROOT))

from absl import app
from absl import flags
from absl import logging
from flax_model.alphagenome._sdk.data import genome
from flax_model.alphagenome._sdk.models import dna_model as dna_model_types
from flax_model.alphagenome._sdk.models import variant_scorers as variant_scorers_lib
import pandas as pd
from flax_model.alphagenome.model.dna_model import (
    OrganismSettings,
    create,
    create_from_kaggle,
)

FLAGS = flags.FLAGS

flags.DEFINE_string(
    "vcf_path",
    None,
    "Path to a VCF file. If unset, built-in demo variants are used.",
)
flags.DEFINE_string(
    "fasta_path",
    None,
    "Path to the reference genome FASTA file. Required when --model_dir is set.",
)
flags.DEFINE_string(
    "model_dir",
    None,
    "Local AlphaGenome checkpoint directory. If unset, Kaggle Hub is used.",
)
flags.DEFINE_string("output_dir", "./outputs", "Directory for CSV outputs.")
flags.DEFINE_enum(
    "organism",
    "HOMO_SAPIENS",
    ["HOMO_SAPIENS", "MUS_MUSCULUS"],
    "Target organism.",
)
flags.DEFINE_enum(
    "model_version",
    "all_folds",
    ["FOLD_0", "FOLD_1", "FOLD_2", "FOLD_3", "FOLD_4", "all_folds"],
    "Model version to download from Kaggle.",
)


DEMO_VARIANTS = [
    ("chr22:36201698:A>C", "eQTL with SuSiE PIP > 0.9 in GTEx Colon"),
    ("chr3:120280774:G>T", "caQTL in GM12878 (DNase)"),
    ("chr21:46126238:G>C", "Splice junction variant in COL6A2"),
]


def load_demo_variants() -> list[tuple[genome.Variant, str]]:
  """Builds the built-in demo variants."""
  return [
      (genome.Variant.from_str(variant_str), description)
      for variant_str, description in DEMO_VARIANTS
  ]


def load_variants_from_vcf(vcf_path: str) -> list[tuple[genome.Variant, str]]:
  """Loads variants from a VCF file."""
  variants_df = pd.read_csv(
      vcf_path,
      sep="\t",
      comment="#",
      names=["CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO"],
  )
  variants_with_desc = []
  for _, row in variants_df.iterrows():
    variant = genome.Variant(
        chromosome=row["CHROM"],
        position=int(row["POS"]),
        reference_bases=row["REF"],
        alternate_bases=str(row["ALT"]).split(",")[0],
    )
    description = row["ID"] if pd.notna(row["ID"]) else "unknown"
    variants_with_desc.append((variant, str(description)))
  return variants_with_desc


def resolve_variant_scorers(
    alphagenome_model,
    organism: dna_model_types.Organism,
) -> tuple[
    Sequence[variant_scorers_lib.VariantScorerTypes],
    list[str],
]:
  """Returns the recommended scorers supported by the current model."""
  recommended_scorers = list(
      variant_scorers_lib.get_recommended_scorers(organism.to_proto())
  )
  available_scorer_map = getattr(alphagenome_model, "_variant_scorers", {}).get(
      organism, {}
  )

  if not available_scorer_map:
    logging.warning(
        "Unable to inspect model variant scorers. Falling back to the full "
        "recommended scorer list."
    )
    return recommended_scorers, []

  available_base_scorers = set(available_scorer_map)
  selected_scorers = [
      scorer
      for scorer in recommended_scorers
      if scorer.base_variant_scorer in available_base_scorers
  ]
  skipped_scorers = [
      scorer.base_variant_scorer.name
      for scorer in recommended_scorers
      if scorer.base_variant_scorer not in available_base_scorers
  ]

  if not selected_scorers:
    available_names = sorted(
        base_scorer.name for base_scorer in available_base_scorers
    )
    raise ValueError(
        "No compatible recommended variant scorers are available for "
        f"{organism.name}. Available scorers: {available_names}."
    )

  return selected_scorers, skipped_scorers


def load_alphagenome_model(organism: dna_model_types.Organism):
  """Loads AlphaGenome from a local checkpoint when provided."""
  organism_settings = None
  if FLAGS.fasta_path:
    organism_settings = {
        organism: OrganismSettings(
            fasta_path=FLAGS.fasta_path,
        ),
    }

  if FLAGS.model_dir:
    if not FLAGS.fasta_path:
      raise ValueError("--fasta_path is required when using --model_dir.")
    return create(
        checkpoint_path=FLAGS.model_dir,
        organism_settings=organism_settings,
    )

  return create_from_kaggle(
      FLAGS.model_version,
      organism_settings=organism_settings,
  )


def main(_):
  output_dir = pathlib.Path(FLAGS.output_dir)
  output_dir.mkdir(parents=True, exist_ok=True)

  organism = dna_model_types.Organism[FLAGS.organism]

  logging.info("Loading AlphaGenome model...")
  alphagenome_model = load_alphagenome_model(organism)

  variant_scorers, skipped_scorers = resolve_variant_scorers(
      alphagenome_model, organism
  )
  logging.info(
      "Using variant scorers: %s",
      ", ".join(scorer.base_variant_scorer.name for scorer in variant_scorers),
  )
  if skipped_scorers:
    logging.warning(
        "Skipping unavailable recommended scorers: %s. This usually means the "
        "model was loaded without the required annotation resources.",
        ", ".join(skipped_scorers),
    )

  if FLAGS.vcf_path:
    logging.info("Loading variants from VCF: %s", FLAGS.vcf_path)
    variants_with_desc = load_variants_from_vcf(FLAGS.vcf_path)
  else:
    logging.info("Using built-in demo variants.")
    variants_with_desc = load_demo_variants()

  logging.info("Scoring %d variants...", len(variants_with_desc))
  all_results = []

  for variant, description in variants_with_desc:
    logging.info("Processing variant: %s (%s)", variant, description)

    interval = variant.reference_interval.resize(2**20)
    scores = alphagenome_model.score_variant(
        interval=interval,
        variant=variant,
        variant_scorers=variant_scorers,
        organism=organism,
    )

    all_results.append(
        {
            "variant": str(variant),
            "description": description,
            "num_score_tables": len(scores),
            "used_variant_scorers": ",".join(
                scorer.base_variant_scorer.name for scorer in variant_scorers
            ),
            "skipped_variant_scorers": ",".join(skipped_scorers),
        }
    )

    for i, adata in enumerate(scores):
      scorer_name = str(adata.uns.get("variant_scorer", f"scorer_{i}"))
      scorer_label = scorer_name.replace(" ", "_").replace("/", "_")
      save_path = output_dir / (
          f"variant_{variant.chromosome}_{variant.position}_{scorer_label}.csv"
      )
      adata.to_df().to_csv(save_path)
      logging.info(
          "Saved score table %d: %s (shape=%s)", i, save_path.name, adata.shape
      )

  summary_path = output_dir / "variant_scoring_summary.csv"
  pd.DataFrame(all_results).to_csv(summary_path, index=False)
  logging.info("Saved summary to %s", summary_path)


if __name__ == "__main__":
  app.run(main)