bio2zarr 0.1.5__py3-none-any.whl → 0.1.7__py3-none-any.whl

bio2zarr/{vcf2zarr/icf.py → vcf.py}

@@ -6,14 +6,19 @@ import logging
 import math
 import pathlib
 import pickle
+import re
 import shutil
 import sys
+import tempfile
+from functools import partial
 from typing import Any
 
 import numcodecs
 import numpy as np
 
-from .. import constants, core, provenance, vcf_utils
+from bio2zarr.zarr_utils import STRING_DTYPE_NAME, zarr_exists
+
+from . import constants, core, provenance, vcf_utils, vcz
 
 logger = logging.getLogger(__name__)
 
@@ -77,6 +82,14 @@ class VcfField:
             return self.name
         return f"{self.category}/{self.name}"
 
+    @property
+    def max_number(self):
+        if self.vcf_number in ("R", "A", "G", "."):
+            return self.summary.max_number
+        else:
+            # use declared number if larger than max found
+            return max(self.summary.max_number, int(self.vcf_number))
+
     def smallest_dtype(self):
         """
         Returns the smallest dtype suitable for this field based
@@ -99,7 +112,7 @@ class VcfField:
             ret = "U1"
         else:
             assert self.vcf_type == "String"
-            ret = "O"
+            ret = STRING_DTYPE_NAME
         return ret
 
 
@@ -116,23 +129,6 @@ ICF_DEFAULT_COMPRESSOR = numcodecs.Blosc(
 )
 
 
-@dataclasses.dataclass
-class Contig:
-    id: str
-    length: int = None
-
-
-@dataclasses.dataclass
-class Sample:
-    id: str
-
-
-@dataclasses.dataclass
-class Filter:
-    id: str
-    description: str = ""
-
-
 @dataclasses.dataclass
 class IcfMetadata(core.JsonDataclass):
     samples: list
@@ -187,9 +183,9 @@ class IcfMetadata(core.JsonDataclass):
         d = d.copy()
         d["partitions"] = partitions
         d["fields"] = [VcfField.fromdict(fd) for fd in d["fields"]]
-        d["samples"] = [Sample(**sd) for sd in d["samples"]]
-        d["filters"] = [Filter(**fd) for fd in d["filters"]]
-        d["contigs"] = [Contig(**cd) for cd in d["contigs"]]
+        d["samples"] = [vcz.Sample(**sd) for sd in d["samples"]]
+        d["filters"] = [vcz.Filter(**fd) for fd in d["filters"]]
+        d["contigs"] = [vcz.Contig(**cd) for cd in d["contigs"]]
         return IcfMetadata(**d)
 
     def __eq__(self, other):
@@ -240,7 +236,7 @@ def scan_vcf(path, target_num_partitions):
                     description = ""
                 if h["ID"] == "PASS":
                     pass_index = len(filters)
-                filters.append(Filter(h["ID"], description))
+                filters.append(vcz.Filter(h["ID"], description))
 
         # Ensure PASS is the first filter if present
         if pass_index > 0:
@@ -262,9 +258,9 @@ def scan_vcf(path, target_num_partitions):
             contig_lengths = [None for _ in vcf.seqnames]
 
         metadata = IcfMetadata(
-            samples=[Sample(sample_id) for sample_id in vcf.samples],
+            samples=[vcz.Sample(sample_id) for sample_id in vcf.samples],
             contigs=[
-                Contig(contig_id, length)
+                vcz.Contig(contig_id, length)
                 for contig_id, length in zip(vcf.seqnames, contig_lengths)
             ],
             filters=filters,
@@ -291,7 +287,12 @@ def scan_vcf(path, target_num_partitions):
         return metadata, vcf.raw_header
 
 
-def scan_vcfs(paths, show_progress, target_num_partitions, worker_processes=1):
+def scan_vcfs(
+    paths,
+    show_progress,
+    target_num_partitions,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
+):
     logger.info(
         f"Scanning {len(paths)} VCFs attempting to split into {target_num_partitions}"
         f" partitions."
@@ -366,64 +367,58 @@ def scan_vcfs(paths, show_progress, target_num_partitions, worker_processes=1):
     return icf_metadata, header
 
 
-def sanitise_value_bool(buff, j, value):
+def sanitise_value_bool(shape, value):
     x = True
     if value is None:
         x = False
-    buff[j] = x
+    return x
 
 
-def sanitise_value_float_scalar(buff, j, value):
+def sanitise_value_float_scalar(shape, value):
     x = value
     if value is None:
         x = [constants.FLOAT32_MISSING]
-    buff[j] = x[0]
+    return x[0]
 
 
-def sanitise_value_int_scalar(buff, j, value):
+def sanitise_value_int_scalar(shape, value):
     x = value
     if value is None:
-        # print("MISSING", INT_MISSING, INT_FILL)
         x = [constants.INT_MISSING]
     else:
         x = sanitise_int_array(value, ndmin=1, dtype=np.int32)
-    buff[j] = x[0]
+    return x[0]
 
 
-def sanitise_value_string_scalar(buff, j, value):
+def sanitise_value_string_scalar(shape, value):
     if value is None:
-        buff[j] = "."
+        return "."
     else:
-        buff[j] = value[0]
+        return value[0]
 
 
-def sanitise_value_string_1d(buff, j, value):
+def sanitise_value_string_1d(shape, value):
     if value is None:
-        buff[j] = "."
+        return np.full(shape, ".", dtype=STRING_DTYPE_NAME)
     else:
-        # value = np.array(value, ndmin=1, dtype=buff.dtype, copy=False)
-        # FIXME failure isn't coming from here, it seems to be from an
-        # incorrectly detected dimension in the zarr array
-        # The dimesions look all wrong, and the dtype should be Object
-        # not str
         value = drop_empty_second_dim(value)
-        buff[j] = ""
-        buff[j, : value.shape[0]] = value
+        result = np.full(shape, "", dtype=value.dtype)
+        result[: value.shape[0]] = value
+        return result
 
 
-def sanitise_value_string_2d(buff, j, value):
+def sanitise_value_string_2d(shape, value):
     if value is None:
-        buff[j] = "."
+        return np.full(shape, ".", dtype=STRING_DTYPE_NAME)
     else:
-        # print(buff.shape, value.dtype, value)
-        # assert value.ndim == 2
-        buff[j] = ""
+        result = np.full(shape, "", dtype=STRING_DTYPE_NAME)
         if value.ndim == 2:
-            buff[j, :, : value.shape[1]] = value
+            result[: value.shape[0], : value.shape[1]] = value
         else:
-            # TODO check if this is still necessary
+            # Convert 1D array into 2D with appropriate shape
             for k, val in enumerate(value):
-                buff[j, k, : len(val)] = val
+                result[k, : len(val)] = val
+        return result
 
 
 def drop_empty_second_dim(value):
@@ -433,27 +428,28 @@ def drop_empty_second_dim(value):
     return value
 
 
-def sanitise_value_float_1d(buff, j, value):
+def sanitise_value_float_1d(shape, value):
     if value is None:
-        buff[j] = constants.FLOAT32_MISSING
+        return np.full(shape, constants.FLOAT32_MISSING)
     else:
-        value = np.array(value, ndmin=1, dtype=buff.dtype, copy=True)
+        value = np.array(value, ndmin=1, dtype=np.float32, copy=True)
         # numpy will map None values to Nan, but we need a
         # specific NaN
         value[np.isnan(value)] = constants.FLOAT32_MISSING
         value = drop_empty_second_dim(value)
-        buff[j] = constants.FLOAT32_FILL
-        buff[j, : value.shape[0]] = value
+        result = np.full(shape, constants.FLOAT32_FILL, dtype=np.float32)
+        result[: value.shape[0]] = value
+        return result
 
 
-def sanitise_value_float_2d(buff, j, value):
+def sanitise_value_float_2d(shape, value):
     if value is None:
-        buff[j] = constants.FLOAT32_MISSING
+        return np.full(shape, constants.FLOAT32_MISSING)
     else:
-        # print("value = ", value)
-        value = np.array(value, ndmin=2, dtype=buff.dtype, copy=True)
-        buff[j] = constants.FLOAT32_FILL
-        buff[j, :, : value.shape[1]] = value
+        value = np.array(value, ndmin=2, dtype=np.float32, copy=True)
+        result = np.full(shape, constants.FLOAT32_FILL, dtype=np.float32)
+        result[:, : value.shape[1]] = value
+        return result
 
 
 def sanitise_int_array(value, ndmin, dtype):
@@ -468,23 +464,25 @@ def sanitise_int_array(value, ndmin, dtype):
     return value.astype(dtype)
 
 
-def sanitise_value_int_1d(buff, j, value):
+def sanitise_value_int_1d(shape, value):
     if value is None:
-        buff[j] = -1
+        return np.full(shape, -1)
     else:
-        value = sanitise_int_array(value, 1, buff.dtype)
+        value = sanitise_int_array(value, 1, np.int32)
         value = drop_empty_second_dim(value)
-        buff[j] = -2
-        buff[j, : value.shape[0]] = value
+        result = np.full(shape, -2, dtype=np.int32)
+        result[: value.shape[0]] = value
+        return result
 
 
-def sanitise_value_int_2d(buff, j, value):
+def sanitise_value_int_2d(shape, value):
     if value is None:
-        buff[j] = -1
+        return np.full(shape, -1)
     else:
-        value = sanitise_int_array(value, 2, buff.dtype)
-        buff[j] = -2
-        buff[j, :, : value.shape[1]] = value
+        value = sanitise_int_array(value, 2, np.int32)
+        result = np.full(shape, -2, dtype=np.int32)
+        result[:, : value.shape[1]] = value
+        return result
 
 
 missing_value_map = {
@@ -573,7 +571,12 @@ class StringValueTransformer(VcfValueTransformer):
             value = np.array(list(vcf_value.split(",")))
         else:
             # TODO can we make this faster??
-            value = np.array([v.split(",") for v in vcf_value], dtype="O")
+            var_len_values = [v.split(",") for v in vcf_value]
+            number = max(len(v) for v in var_len_values)
+            value = np.array(
+                [v + [""] * (number - len(v)) for v in var_len_values],
+                dtype=STRING_DTYPE_NAME,
+            )
             # print("HERE", vcf_value, value)
             # for v in vcf_value:
             #     print("\t", type(v), len(v), v.split(","))
@@ -648,7 +651,8 @@ class IntermediateColumnarFormatField:
         chunk_cumulative_records = self.chunk_record_index(partition_id)
         chunk_num_records = np.diff(chunk_cumulative_records)
         for count, cumulative in zip(
-            chunk_num_records[start_chunk:], chunk_cumulative_records[start_chunk + 1 :]
+            chunk_num_records[start_chunk:],
+            chunk_cumulative_records[start_chunk + 1 :],
         ):
             path = partition_path / f"{cumulative}"
             chunk = self.read_chunk(path)
@@ -707,36 +711,32 @@ class IntermediateColumnarFormatField:
         return ret
 
     def sanitiser_factory(self, shape):
-        """
-        Return a function that sanitised values from this column
-        and writes into a buffer of the specified shape.
-        """
-        assert len(shape) <= 3
+        assert len(shape) <= 2
         if self.vcf_field.vcf_type == "Flag":
-            assert len(shape) == 1
-            return sanitise_value_bool
+            assert len(shape) == 0
+            return partial(sanitise_value_bool, shape)
         elif self.vcf_field.vcf_type == "Float":
-            if len(shape) == 1:
-                return sanitise_value_float_scalar
-            elif len(shape) == 2:
-                return sanitise_value_float_1d
+            if len(shape) == 0:
+                return partial(sanitise_value_float_scalar, shape)
+            elif len(shape) == 1:
+                return partial(sanitise_value_float_1d, shape)
             else:
-                return sanitise_value_float_2d
+                return partial(sanitise_value_float_2d, shape)
         elif self.vcf_field.vcf_type == "Integer":
-            if len(shape) == 1:
-                return sanitise_value_int_scalar
-            elif len(shape) == 2:
-                return sanitise_value_int_1d
+            if len(shape) == 0:
+                return partial(sanitise_value_int_scalar, shape)
+            elif len(shape) == 1:
+                return partial(sanitise_value_int_1d, shape)
             else:
-                return sanitise_value_int_2d
+                return partial(sanitise_value_int_2d, shape)
         else:
             assert self.vcf_field.vcf_type in ("String", "Character")
-            if len(shape) == 1:
-                return sanitise_value_string_scalar
-            elif len(shape) == 2:
-                return sanitise_value_string_1d
+            if len(shape) == 0:
+                return partial(sanitise_value_string_scalar, shape)
+            elif len(shape) == 1:
+                return partial(sanitise_value_string_1d, shape)
             else:
-                return sanitise_value_string_2d
+                return partial(sanitise_value_string_2d, shape)
 
 
 @dataclasses.dataclass
@@ -843,9 +843,66 @@ class IcfPartitionWriter(contextlib.AbstractContextManager):
         return False
 
 
-class IntermediateColumnarFormat(collections.abc.Mapping):
+def convert_local_allele_field_types(fields, schema_instance):
+    """
+    Update the specified list of fields to include the LAA field, and to convert
+    any supported localisable fields to the L* counterpart.
+
+    Note that we currently support only two ALT alleles per sample, and so the
+    dimensions of these fields are fixed by that requirement. Later versions may
+    use summary data storted in the ICF to make different choices, if information
+    about subsequent alleles (not in the actual genotype calls) should also be
+    stored.
+    """
+    fields_by_name = {field.name: field for field in fields}
+    gt = fields_by_name["call_genotype"]
+
+    if schema_instance.get_shape(["ploidy"])[0] != 2:
+        raise ValueError("Local alleles only supported on diploid data")
+
+    dimensions = gt.dimensions[:-1]
+
+    la = vcz.ZarrArraySpec(
+        name="call_LA",
+        dtype="i1",
+        dimensions=(*dimensions, "local_alleles"),
+        description=(
+            "0-based indices into REF+ALT, indicating which alleles"
+            " are relevant (local) for the current sample"
+        ),
+    )
+    schema_instance.dimensions["local_alleles"] = vcz.VcfZarrDimension.unchunked(
+        schema_instance.dimensions["ploidy"].size
+    )
+
+    ad = fields_by_name.get("call_AD", None)
+    if ad is not None:
+        # TODO check if call_LAD is in the list already
+        ad.name = "call_LAD"
+        ad.source = None
+        ad.dimensions = (*dimensions, "local_alleles_AD")
+        ad.description += " (local-alleles)"
+        schema_instance.dimensions["local_alleles_AD"] = vcz.VcfZarrDimension.unchunked(
+            2
+        )
+
+    pl = fields_by_name.get("call_PL", None)
+    if pl is not None:
+        # TODO check if call_LPL is in the list already
+        pl.name = "call_LPL"
+        pl.source = None
+        pl.description += " (local-alleles)"
+        pl.dimensions = (*dimensions, "local_" + pl.dimensions[-1].split("_")[-1])
+        schema_instance.dimensions["local_" + pl.dimensions[-1].split("_")[-1]] = (
+            vcz.VcfZarrDimension.unchunked(3)
+        )
+
+    return [*fields, la]
+
+
+class IntermediateColumnarFormat(vcz.Source):
     def __init__(self, path):
-        self.path = pathlib.Path(path)
+        self._path = pathlib.Path(path)
         # TODO raise a more informative error here telling people this
         # directory is either a WIP or the wrong format.
         with open(self.path / "metadata.json") as f:
@@ -859,8 +916,12 @@ class IntermediateColumnarFormat(collections.abc.Mapping):
         ]
         # Allow us to find which partition a given record is in
         self.partition_record_index = np.cumsum([0, *partition_num_records])
+        self.gt_field = None
         for field in self.metadata.fields:
             self.fields[field.full_name] = IntermediateColumnarFormatField(self, field)
+            if field.name == "GT":
+                self.gt_field = field
+
         logger.info(
             f"Loaded IntermediateColumnarFormat(partitions={self.num_partitions}, "
             f"records={self.num_records}, fields={self.num_fields})"
@@ -868,20 +929,11 @@ class IntermediateColumnarFormat(collections.abc.Mapping):
 
     def __repr__(self):
         return (
-            f"IntermediateColumnarFormat(fields={len(self)}, "
+            f"IntermediateColumnarFormat(fields={len(self.fields)}, "
             f"partitions={self.num_partitions}, "
             f"records={self.num_records}, path={self.path})"
         )
 
-    def __getitem__(self, key):
-        return self.fields[key]
-
-    def __iter__(self):
-        return iter(self.fields)
-
-    def __len__(self):
-        return len(self.fields)
-
     def summary_table(self):
         data = []
         for name, icf_field in self.fields.items():
@@ -900,6 +952,10 @@ class IntermediateColumnarFormat(collections.abc.Mapping):
             data.append(d)
         return data
 
+    @property
+    def path(self):
+        return self._path
+
     @property
     def num_records(self):
         return self.metadata.num_records
@@ -908,6 +964,18 @@ class IntermediateColumnarFormat(collections.abc.Mapping):
     def num_partitions(self):
         return len(self.metadata.partitions)
 
+    @property
+    def samples(self):
+        return self.metadata.samples
+
+    @property
+    def contigs(self):
+        return self.metadata.contigs
+
+    @property
+    def filters(self):
+        return self.metadata.filters
+
     @property
     def num_samples(self):
         return len(self.metadata.samples)
@@ -916,6 +984,265 @@ class IntermediateColumnarFormat(collections.abc.Mapping):
     def num_fields(self):
         return len(self.fields)
 
+    @property
+    def root_attrs(self):
+        meta_information_pattern = re.compile("##([^=]+)=(.*)")
+        vcf_meta_information = []
+        for line in self.vcf_header.split("\n"):
+            match = re.fullmatch(meta_information_pattern, line)
+            if match:
+                key = match.group(1)
+                if key in ("contig", "FILTER", "INFO", "FORMAT"):
+                    # these fields are stored in Zarr arrays
+                    continue
+                value = match.group(2)
+                vcf_meta_information.append((key, value))
+        return {
+            "vcf_meta_information": vcf_meta_information,
+        }
+
+    def iter_id(self, start, stop):
+        for value in self.fields["ID"].iter_values(start, stop):
+            if value is not None:
+                yield value[0]
+            else:
+                yield None
+
+    def iter_filters(self, start, stop):
+        source_field = self.fields["FILTERS"]
+        lookup = {filt.id: index for index, filt in enumerate(self.metadata.filters)}
+
+        for filter_values in source_field.iter_values(start, stop):
+            filters = np.zeros(len(self.metadata.filters), dtype=bool)
+            if filter_values is not None:
+                for filter_id in filter_values:
+                    try:
+                        filters[lookup[filter_id]] = True
+                    except KeyError:
+                        raise ValueError(
+                            f"Filter '{filter_id}' was not defined in the header."
+                        ) from None
+            yield filters
+
+    def iter_contig(self, start, stop):
+        source_field = self.fields["CHROM"]
+        lookup = {
+            contig.id: index for index, contig in enumerate(self.metadata.contigs)
+        }
+
+        for value in source_field.iter_values(start, stop):
+            # Note: because we are using the indexes to define the lookups
+            # and we always have an index, it seems that we the contig lookup
+            # will always succeed. However, if anyone ever does hit a KeyError
+            # here, please do open an issue with a reproducible example!
+            yield lookup[value[0]]
+
+    def iter_field(self, field_name, shape, start, stop):
+        source_field = self.fields[field_name]
+        sanitiser = source_field.sanitiser_factory(shape)
+        for value in source_field.iter_values(start, stop):
+            yield sanitiser(value)
+
+    def iter_alleles(self, start, stop, num_alleles):
+        ref_field = self.fields["REF"]
+        alt_field = self.fields["ALT"]
+
+        for ref, alt in zip(
+            ref_field.iter_values(start, stop),
+            alt_field.iter_values(start, stop),
+        ):
+            alleles = np.full(num_alleles, constants.STR_FILL, dtype=STRING_DTYPE_NAME)
+            alleles[0] = ref[0]
+            alleles[1 : 1 + len(alt)] = alt
+            yield alleles
+
+    def iter_genotypes(self, shape, start, stop):
+        source_field = self.fields["FORMAT/GT"]
+        for value in source_field.iter_values(start, stop):
+            genotypes = value[:, :-1] if value is not None else None
+            phased = value[:, -1] if value is not None else None
+            sanitised_genotypes = sanitise_value_int_2d(shape, genotypes)
+            sanitised_phased = sanitise_value_int_1d(shape[:-1], phased)
+            # Force haploids to always be phased
+            # https://github.com/sgkit-dev/bio2zarr/issues/399
+            if sanitised_genotypes.shape[1] == 1:
+                sanitised_phased[:] = True
+            yield sanitised_genotypes, sanitised_phased
+
+    def iter_alleles_and_genotypes(self, start, stop, shape, num_alleles):
+        variant_lengths = self.fields["rlen"].iter_values(start, stop)
+        if self.gt_field is None or shape is None:
+            for variant_length, alleles in zip(
+                variant_lengths, self.iter_alleles(start, stop, num_alleles)
+            ):
+                # Stored ICF values are always at least 1D arrays; "rlen" is Number=1
+                # so we must extract the scalar to avoid NumPy scalar-conversion issues.
+                yield vcz.VariantData(variant_length[0], alleles, None, None)
+        else:
+            for variant_length, alleles, (gt, phased) in zip(
+                variant_lengths,
+                self.iter_alleles(start, stop, num_alleles),
+                self.iter_genotypes(shape, start, stop),
+            ):
+                yield vcz.VariantData(variant_length[0], alleles, gt, phased)
+
+    def generate_schema(
+        self, variants_chunk_size=None, samples_chunk_size=None, local_alleles=None
+    ):
+        if local_alleles is None:
+            local_alleles = False
+
+        max_alleles = max(self.fields["ALT"].vcf_field.summary.max_number + 1, 2)
+
+        # Add ploidy and genotypes dimensions only when needed
+        max_genotypes = 0
+        has_g_field = False
+        for field in self.metadata.format_fields:
+            if field.vcf_number == "G":
+                has_g_field = True
+                max_genotypes = max(max_genotypes, field.summary.max_number)
+
+        ploidy = None
+        genotypes_size = None
+        if self.gt_field is not None:
+            ploidy = max(self.gt_field.summary.max_number - 1, 1)
+            # NOTE: it's not clear why we're computing this, when we must have had
+            # at least one number=G field to require it anyway?
+            genotypes_size = math.comb(max_alleles + ploidy - 1, ploidy)
+            # assert max_genotypes == genotypes_size
+        else:
+            if max_genotypes > 0 or has_g_field:
+                # there is no GT field, but there is at least one Number=G field,
+                # so need to define genotypes dimension
+                genotypes_size = max_genotypes
+
+        dimensions = vcz.standard_dimensions(
+            variants_size=self.num_records,
+            variants_chunk_size=variants_chunk_size,
+            samples_size=self.num_samples,
+            samples_chunk_size=samples_chunk_size,
+            alleles_size=max_alleles,
+            filters_size=self.metadata.num_filters,
+            ploidy_size=ploidy,
+            genotypes_size=genotypes_size,
+        )
+
+        schema_instance = vcz.VcfZarrSchema(
+            format_version=vcz.ZARR_SCHEMA_FORMAT_VERSION,
+            dimensions=dimensions,
+            fields=[],
+        )
+
+        logger.info(
+            "Generating schema with chunks="
+            f"variants={dimensions['variants'].chunk_size}, "
+            f"samples={dimensions['samples'].chunk_size}"
+        )
+
+        def spec_from_field(field, array_name=None):
+            return vcz.ZarrArraySpec.from_field(
+                field,
+                schema_instance,
+                array_name=array_name,
+            )
+
+        def fixed_field_spec(name, dtype, source=None, dimensions=("variants",)):
+            compressor = (
+                vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config()
+                if dtype == "bool"
+                else None
+            )
+            return vcz.ZarrArraySpec(
+                source=source,
+                name=name,
+                dtype=dtype,
+                description="",
+                dimensions=dimensions,
+                compressor=compressor,
+            )
+
+        name_map = {field.full_name: field for field in self.metadata.fields}
+        array_specs = [
+            fixed_field_spec(
+                name="variant_contig",
+                dtype=core.min_int_dtype(0, self.metadata.num_contigs),
+            ),
+            fixed_field_spec(
+                name="variant_filter",
+                dtype="bool",
+                dimensions=["variants", "filters"],
+            ),
+            fixed_field_spec(
+                name="variant_allele",
+                dtype=STRING_DTYPE_NAME,
+                dimensions=["variants", "alleles"],
+            ),
+            fixed_field_spec(
+                name="variant_length",
+                dtype=name_map["rlen"].smallest_dtype(),
+                dimensions=["variants"],
+            ),
+            fixed_field_spec(
+                name="variant_id",
+                dtype=STRING_DTYPE_NAME,
+            ),
+            fixed_field_spec(
+                name="variant_id_mask",
+                dtype="bool",
+            ),
+        ]
+
+        # Only two of the fixed fields have a direct one-to-one mapping.
+        array_specs.extend(
+            [
+                spec_from_field(name_map["QUAL"], array_name="variant_quality"),
+                spec_from_field(name_map["POS"], array_name="variant_position"),
+            ]
+        )
+        array_specs.extend(
+            [spec_from_field(field) for field in self.metadata.info_fields]
+        )
+
+        for field in self.metadata.format_fields:
+            if field.name == "GT":
+                continue
+            array_specs.append(spec_from_field(field))
+
+        if self.gt_field is not None and self.num_samples > 0:
+            array_specs.append(
+                vcz.ZarrArraySpec(
+                    name="call_genotype_phased",
+                    dtype="bool",
+                    dimensions=["variants", "samples"],
+                    description="",
+                    compressor=vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config(),
+                )
+            )
+            array_specs.append(
+                vcz.ZarrArraySpec(
+                    name="call_genotype",
+                    dtype=self.gt_field.smallest_dtype(),
+                    dimensions=["variants", "samples", "ploidy"],
+                    description="",
+                    compressor=vcz.DEFAULT_ZARR_COMPRESSOR_GENOTYPES.get_config(),
+                )
+            )
+            array_specs.append(
+                vcz.ZarrArraySpec(
+                    name="call_genotype_mask",
+                    dtype="bool",
+                    dimensions=["variants", "samples", "ploidy"],
+                    description="",
+                    compressor=vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config(),
+                )
+            )
+
+        if local_alleles:
+            array_specs = convert_local_allele_field_types(array_specs, schema_instance)
+
+        schema_instance.fields = array_specs
+        return schema_instance
+
 
 @dataclasses.dataclass
 class IcfPartitionMetadata(core.JsonDataclass):
@@ -987,7 +1314,7 @@ class IntermediateColumnarFormatWriter:
         vcfs,
         *,
         column_chunk_size=16,
-        worker_processes=1,
+        worker_processes=core.DEFAULT_WORKER_PROCESSES,
         target_num_partitions=None,
         show_progress=False,
         compressor=None,
@@ -1139,7 +1466,9 @@ class IntermediateColumnarFormatWriter:
             f"{num_records} records last_pos={last_position}"
         )
 
-    def explode(self, *, worker_processes=1, show_progress=False):
+    def explode(
+        self, *, worker_processes=core.DEFAULT_WORKER_PROCESSES, show_progress=False
+    ):
         self.load_metadata()
         num_records = self.metadata.num_records
         if np.isinf(num_records):
@@ -1207,7 +1536,7 @@ def explode(
     vcfs,
     *,
     column_chunk_size=16,
-    worker_processes=1,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
     show_progress=False,
     compressor=None,
 ):
@@ -1232,7 +1561,7 @@ def explode_init(
     *,
     column_chunk_size=16,
     target_num_partitions=1,
-    worker_processes=1,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
     show_progress=False,
     compressor=None,
 ):
@@ -1255,3 +1584,166 @@ def explode_partition(icf_path, partition):
 def explode_finalise(icf_path):
     writer = IntermediateColumnarFormatWriter(icf_path)
     writer.finalise()
+
+
+def inspect(path):
+    path = pathlib.Path(path)
+    if not path.exists():
+        raise ValueError(f"Path not found: {path}")
+    if (path / "metadata.json").exists():
+        obj = IntermediateColumnarFormat(path)
+    elif zarr_exists(path):
+        obj = vcz.VcfZarr(path)
+    else:
+        raise ValueError(f"{path} not in ICF or VCF Zarr format")
+    return obj.summary_table()
+
+
+def mkschema(
+    if_path,
+    out,
+    *,
+    variants_chunk_size=None,
+    samples_chunk_size=None,
+    local_alleles=None,
+):
+    store = IntermediateColumnarFormat(if_path)
+    spec = store.generate_schema(
+        variants_chunk_size=variants_chunk_size,
+        samples_chunk_size=samples_chunk_size,
+        local_alleles=local_alleles,
+    )
+    out.write(spec.asjson())
+
+
+def convert(
+    vcfs,
+    vcz_path,
+    *,
+    variants_chunk_size=None,
+    samples_chunk_size=None,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
+    local_alleles=None,
+    show_progress=False,
+    icf_path=None,
+):
+    """
+    Convert the VCF data at the specified list of paths
+    to VCF Zarr format stored at the specified path.
+
+    .. todo:: Document parameters
+    """
+    if icf_path is None:
+        cm = temp_icf_path(prefix="vcf2zarr")
+    else:
+        cm = contextlib.nullcontext(icf_path)
+
+    with cm as icf_path:
+        explode(
+            icf_path,
+            vcfs,
+            worker_processes=worker_processes,
+            show_progress=show_progress,
+        )
+        encode(
+            icf_path,
+            vcz_path,
+            variants_chunk_size=variants_chunk_size,
+            samples_chunk_size=samples_chunk_size,
+            worker_processes=worker_processes,
+            show_progress=show_progress,
+            local_alleles=local_alleles,
+        )
+
+
+@contextlib.contextmanager
+def temp_icf_path(prefix=None):
+    with tempfile.TemporaryDirectory(prefix=prefix) as tmp:
+        yield pathlib.Path(tmp) / "icf"
+
+
+def encode(
+    icf_path,
+    zarr_path,
+    schema_path=None,
+    variants_chunk_size=None,
+    samples_chunk_size=None,
+    max_variant_chunks=None,
+    dimension_separator=None,
+    max_memory=None,
+    local_alleles=None,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
+    show_progress=False,
+):
+    # Rough heuristic to split work up enough to keep utilisation high
+    target_num_partitions = max(1, worker_processes * 4)
+    encode_init(
+        icf_path,
+        zarr_path,
+        target_num_partitions,
+        schema_path=schema_path,
+        variants_chunk_size=variants_chunk_size,
+        samples_chunk_size=samples_chunk_size,
+        local_alleles=local_alleles,
+        max_variant_chunks=max_variant_chunks,
+        dimension_separator=dimension_separator,
+    )
+    vzw = vcz.VcfZarrWriter(IntermediateColumnarFormat, zarr_path)
+    vzw.encode_all_partitions(
+        worker_processes=worker_processes,
+        show_progress=show_progress,
+        max_memory=max_memory,
+    )
+    vzw.finalise(show_progress)
+    vzw.create_index()
+
+
+def encode_init(
+    icf_path,
+    zarr_path,
+    target_num_partitions,
+    *,
+    schema_path=None,
+    variants_chunk_size=None,
+    samples_chunk_size=None,
+    local_alleles=None,
+    max_variant_chunks=None,
+    dimension_separator=None,
+    max_memory=None,
+    worker_processes=core.DEFAULT_WORKER_PROCESSES,
+    show_progress=False,
+):
+    icf_store = IntermediateColumnarFormat(icf_path)
+    if schema_path is None:
+        schema_instance = icf_store.generate_schema(
+            variants_chunk_size=variants_chunk_size,
+            samples_chunk_size=samples_chunk_size,
+            local_alleles=local_alleles,
+        )
+    else:
+        logger.info(f"Reading schema from {schema_path}")
+        if variants_chunk_size is not None or samples_chunk_size is not None:
+            raise ValueError(
+                "Cannot specify schema along with chunk sizes"
+            )  # NEEDS TEST
+        with open(schema_path) as f:
+            schema_instance = vcz.VcfZarrSchema.fromjson(f.read())
+    zarr_path = pathlib.Path(zarr_path)
+    vzw = vcz.VcfZarrWriter("icf", zarr_path)
+    return vzw.init(
+        icf_store,
+        target_num_partitions=target_num_partitions,
+        schema=schema_instance,
+        dimension_separator=dimension_separator,
+        max_variant_chunks=max_variant_chunks,
+    )
+
+
+def encode_partition(zarr_path, partition):
+    writer_instance = vcz.VcfZarrWriter(IntermediateColumnarFormat, zarr_path)
+    writer_instance.encode_partition(partition)
+
+
+def encode_finalise(zarr_path, show_progress=False):
+    writer_instance = vcz.VcfZarrWriter(IntermediateColumnarFormat, zarr_path)
+    writer_instance.finalise(show_progress=show_progress)