cov-unibind.py

import pandas as pd
import datasets

_DESCRIPTION = """\
Multi-source dataset of antibody-mutation interactions including IC50, binding, escape, and affinity measurements.
Also includes antibody synonyms with CDR sequences and epitope information.
"""

_FEATURES = {
    'antibody_name': datasets.Value("string"),
    'antigen_lineage': datasets.Value("string"),
    'target_value': datasets.Value("float"),
    'target_type': datasets.Value("string"),
    'source_name': datasets.Value("string"),
    'source_doi': datasets.Value("string"),
    'assay_name': datasets.Value("string"),
    'pdb_id': datasets.Value("string"),
    'structure_release_date': datasets.Value("string"),
    'structure_resolution': datasets.Value("float"),
    'mutations': datasets.Value("string"),
    'antigen_chain_ids': datasets.Value("string"),
    'antigen_domain': datasets.Value("string"),
    'antigen_residue_indices': datasets.Value("string"),
    'antigen_residue_indices_trimmed': datasets.Value("string"),
    'antigen_host': datasets.Value("string"),
    'antibody_heavy_chain_id': datasets.Value("string"),
    'antibody_light_chain_id': datasets.Value("string"),
    'epitope_residues': datasets.Value("string"),
    'epitope_mutations': datasets.Value("string"),
    'epitope_domain': datasets.Value("string"),
    'epitope_alteration_count': datasets.Value("string"),
    'spike_sequence': datasets.Value("string"),
    'antibody_heavy_chain_sequence': datasets.Value("string"),
    'antibody_light_chain_sequence': datasets.Value("string"),
    'antibody_vh_sequence': datasets.Value("string"),
    'antibody_vl_sequence': datasets.Value("string"),
    'antigen_sequence': datasets.Value("string"),
    'antigen_sequence_trimmed': datasets.Value("string"),
    'antigen_sequence_without_indels': datasets.Value("string"),
    'antigen_sequence_trimmed_without_indels': datasets.Value("string"),
    'antigen_pdb_sequence': datasets.Value("string"),
    'antigen_pdb_sequence_trimmed': datasets.Value("string"),
}

_ANTIBODY_SYNONYMS_FEATURES = {
    'antibody_name': datasets.Value("string"),
    'pdb_id': datasets.Value("string"),
    'antibody_heavy_chain_cdr1': datasets.Value("string"),
    'antibody_heavy_chain_cdr2': datasets.Value("string"),
    'antibody_heavy_chain_cdr3': datasets.Value("string"),
    'antibody_light_chain_cdr1': datasets.Value("string"),
    'antibody_light_chain_cdr2': datasets.Value("string"),
    'antibody_light_chain_cdr3': datasets.Value("string"),
    'epitope_residues': datasets.Value("string"),
    'epitope_domain': datasets.Value("string"),
}

_TABLES = {
    "antibody_synonyms": {
        "file": "antibody_info/antibody_synonyms_with_epitopes.parquet",
        "features": _ANTIBODY_SYNONYMS_FEATURES,
    },
    "drdb": {
        "file": "data/drdb_binding_potency.parquet",
        "features": {
            **_FEATURES,
        }
    },
    "covabdab": {
        "file": "data/covabdab_binding.parquet",
        "features": {
            **{
                **_FEATURES,
                "target_value": datasets.Value("bool"),
            }
        }
    },
    "dms_bloom": {
        "file": "data/dms_bloom_ab_escape.parquet",
        "features": {
            **_FEATURES,
        }
    },
    "dms_cao": {
        "file": "data/dms_cao_ab_escape.parquet",
        "features": {
            **_FEATURES,
        }
    },
    "jian_elisa": {
        "file": "data/jian_elisa_ab_ic50.parquet",
        "features": {
            **_FEATURES,
        }
    },
    "spr": {
        "file": "data/spr_ab_affinity.parquet",
        "features": {
            **_FEATURES,
        }
    }
}

class CovUniBindConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super().__init__(version=datasets.Version("1.0.0"), **kwargs)


class CovUniBind(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        CovUniBindConfig(name=table, description=f"{table} subset") for table in _TABLES
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(_TABLES[self.config.name]["features"]),
        )

    def _split_generators(self, dl_manager):
        file_path = _TABLES[self.config.name]["file"]
        data_path = dl_manager.download_and_extract(file_path)
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_path}),
        ]

    def _generate_examples(self, filepath):
        df = pd.read_parquet(filepath)
        for idx, row in df.iterrows():
            yield idx, row.to_dict()