Inital ts convert

benjeffery · benjeffery · commit 0c6fd5ecf798 · 2025-04-30T01:16:21.000+01:00
diff --git a/bio2zarr/plink.py b/bio2zarr/plink.py
@@ -113,7 +113,7 @@ def generate_schema(
                 dtype="i1",
                 dimensions=["variants", "samples", "ploidy"],
                 description=None,
-                compressor=vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config(),
+                compressor=vcz.DEFAULT_ZARR_COMPRESSOR_GENOTYPES.get_config(),
             ),
             vcz.ZarrArraySpec(
                 name="call_genotype_mask",
diff --git a/bio2zarr/ts.py b/bio2zarr/ts.py
@@ -0,0 +1,287 @@
+import logging
+import pathlib
+
+import numpy as np
+import tskit
+
+from bio2zarr import constants, core, vcz
+
+logger = logging.getLogger(__name__)
+
+
+class TskitFormat(vcz.Source):
+    def __init__(self, ts_path, contig_id=None, ploidy=None, isolated_as_missing=False):
+        self._path = ts_path
+        self.ts = tskit.load(ts_path)
+        self.contig_id = contig_id if contig_id is not None else "1"
+        self.isolated_as_missing = isolated_as_missing
+
+        self._make_sample_mapping(ploidy)
+        self.positions = self.ts.sites_position
+
+    @property
+    def path(self):
+        return self._path
+
+    @property
+    def num_records(self):
+        return self.ts.num_sites
+
+    @property
+    def num_samples(self):
+        return self._num_samples
+
+    @property
+    def samples(self):
+        return self._samples
+
+    @property
+    def root_attrs(self):
+        return {}
+
+    @property
+    def contigs(self):
+        return [vcz.Contig(id=self.contig_id)]
+
+    def _make_sample_mapping(self, ploidy):
+        ts = self.ts
+        self.individual_ploidies = []
+        self.max_ploidy = 0
+
+        if ts.num_individuals > 0 and ploidy is not None:
+            raise ValueError(
+                "Cannot specify ploidy when individuals are present in tables"
+            )
+
+        # Find all sample nodes that reference individuals
+        individuals = np.unique(ts.nodes_individual[ts.samples()])
+        if len(individuals) == 1 and individuals[0] == tskit.NULL:
+            # No samples refer to individuals
+            individuals = None
+        else:
+            # np.unique sorts the argument, so if NULL (-1) is present it
+            # will be the first value.
+            if individuals[0] == tskit.NULL:
+                raise ValueError(
+                    "Sample nodes must either all be associated with individuals "
+                    "or not associated with any individuals"
+                )
+
+        if individuals is not None:
+            self.sample_ids = []
+            for i in individuals:
+                if i < 0 or i >= self.ts.num_individuals:
+                    raise ValueError("Invalid individual IDs provided.")
+                ind = self.ts.individual(i)
+                if len(ind.nodes) == 0:
+                    raise ValueError(f"Individual {i} not associated with a node")
+                is_sample = {ts.node(u).is_sample() for u in ind.nodes}
+                if len(is_sample) != 1:
+                    raise ValueError(
+                        f"Individual {ind.id} has nodes that are sample and "
+                        "non-samples"
+                    )
+                self.sample_ids.extend(ind.nodes)
+                self.individual_ploidies.append(len(ind.nodes))
+                self.max_ploidy = max(self.max_ploidy, len(ind.nodes))
+        else:
+            if ploidy is None:
+                ploidy = 1
+            if ploidy < 1:
+                raise ValueError("Ploidy must be >= 1")
+            if ts.num_samples % ploidy != 0:
+                raise ValueError("Sample size must be divisible by ploidy")
+            self.individual_ploidies = np.full(
+                ts.num_samples // ploidy, ploidy, dtype=np.int32
+            )
+            self.max_ploidy = ploidy
+            self.sample_ids = np.arange(ts.num_samples, dtype=np.int32)
+
+        self._num_samples = len(self.individual_ploidies)
+
+        self._samples = [vcz.Sample(id=f"tsk_{j}") for j in range(self.num_samples)]
+
+    def iter_contig(self, start, stop):
+        yield from (0 for _ in range(start, stop))
+
+    def iter_field(self, field_name, shape, start, stop):
+        if field_name == "position":
+            for pos in self.ts.sites_position[start:stop]:
+                yield int(pos)
+        else:
+            raise ValueError(f"Unknown field {field_name}")
+
+    def iter_alleles(self, start, stop, num_alleles):
+        for variant in self.ts.variants(
+            samples=self.sample_ids,
+            isolated_as_missing=self.isolated_as_missing,
+            left=self.positions[start],
+            right=self.positions[stop] if stop < self.num_records else None,
+        ):
+            alleles = np.full(num_alleles, constants.STR_FILL, dtype="O")
+            for i, allele in enumerate(variant.alleles):
+                assert i < num_alleles
+                alleles[i] = allele
+            yield alleles
+
+    def iter_alleles_and_genotypes(self, start, stop, shape, num_alleles):
+        gt = np.zeros(shape, dtype=np.int8)
+        phased = np.zeros(shape[:-1], dtype=bool)
+
+        for variant in self.ts.variants(
+            samples=self.sample_ids,
+            isolated_as_missing=self.isolated_as_missing,
+            left=self.positions[start],
+            right=self.positions[stop] if stop < self.num_records else None,
+        ):
+            alleles = np.full(num_alleles, constants.STR_FILL, dtype="O")
+            for i, allele in enumerate(variant.alleles):
+                assert i < num_alleles
+                alleles[i] = allele
+
+            genotypes = variant.genotypes
+            sample_index = 0
+            for i, ploidy in enumerate(self.individual_ploidies):
+                for j in range(ploidy):
+                    if j < self.max_ploidy:  # Only fill up to max_ploidy
+                        try:
+                            gt[i, j] = genotypes[sample_index + j]
+                        except IndexError:
+                            # This can happen if the ploidy varies between individuals
+                            gt[i, j] = -2  # Fill value
+
+                # In tskit, all genotypes are considered phased
+                phased[i] = True
+                sample_index += ploidy
+
+            yield alleles, (gt, phased)
+
+    def generate_schema(
+        self,
+        variants_chunk_size=None,
+        samples_chunk_size=None,
+    ):
+        n = self.num_samples
+        m = self.ts.num_sites
+
+        # Determine max number of alleles
+        max_alleles = 0
+        for variant in self.ts.variants():
+            max_alleles = max(max_alleles, len(variant.alleles))
+
+        logging.info(f"Scanned tskit with {n} samples and {m} variants")
+        logging.info(
+            f"Maximum ploidy: {self.max_ploidy}, maximum alleles: {max_alleles}"
+        )
+
+        dimensions = {
+            "variants": vcz.VcfZarrDimension(
+                size=m, chunk_size=variants_chunk_size or vcz.DEFAULT_VARIANT_CHUNK_SIZE
+            ),
+            "samples": vcz.VcfZarrDimension(
+                size=n, chunk_size=samples_chunk_size or vcz.DEFAULT_SAMPLE_CHUNK_SIZE
+            ),
+            "ploidy": vcz.VcfZarrDimension(size=self.max_ploidy),
+            "alleles": vcz.VcfZarrDimension(size=max_alleles),
+        }
+
+        schema_instance = vcz.VcfZarrSchema(
+            format_version=vcz.ZARR_SCHEMA_FORMAT_VERSION,
+            dimensions=dimensions,
+            fields=[],
+        )
+
+        logger.info(
+            "Generating schema with chunks="
+            f"{schema_instance.dimensions['variants'].chunk_size}, "
+            f"{schema_instance.dimensions['samples'].chunk_size}"
+        )
+
+        array_specs = [
+            vcz.ZarrArraySpec(
+                source="position",
+                name="variant_position",
+                dtype="i4",
+                dimensions=["variants"],
+                description="Position of each variant",
+            ),
+            vcz.ZarrArraySpec(
+                source=None,
+                name="variant_allele",
+                dtype="O",
+                dimensions=["variants", "alleles"],
+                description="Alleles for each variant",
+            ),
+            vcz.ZarrArraySpec(
+                source=None,
+                name="variant_contig",
+                dtype=core.min_int_dtype(0, len(self.contigs)),
+                dimensions=["variants"],
+                description="Contig/chromosome index for each variant",
+            ),
+            vcz.ZarrArraySpec(
+                source=None,
+                name="call_genotype_phased",
+                dtype="bool",
+                dimensions=["variants", "samples"],
+                description="Whether the genotype is phased",
+                compressor=vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config(),
+            ),
+            vcz.ZarrArraySpec(
+                source=None,
+                name="call_genotype",
+                dtype="i1",
+                dimensions=["variants", "samples", "ploidy"],
+                description="Genotype for each variant and sample",
+                compressor=vcz.DEFAULT_ZARR_COMPRESSOR_GENOTYPES.get_config(),
+            ),
+            vcz.ZarrArraySpec(
+                source=None,
+                name="call_genotype_mask",
+                dtype="bool",
+                dimensions=["variants", "samples", "ploidy"],
+                description="Mask for each genotype call",
+                compressor=vcz.DEFAULT_ZARR_COMPRESSOR_BOOL.get_config(),
+            ),
+        ]
+        schema_instance.fields = array_specs
+        return schema_instance
+
+
+def convert(
+    ts_path,
+    zarr_path,
+    *,
+    contig_id=None,
+    ploidy=None,
+    isolated_as_missing=False,
+    variants_chunk_size=None,
+    samples_chunk_size=None,
+    worker_processes=1,
+    show_progress=False,
+):
+    tskit_format = TskitFormat(
+        ts_path,
+        contig_id=contig_id,
+        ploidy=ploidy,
+        isolated_as_missing=isolated_as_missing,
+    )
+    schema_instance = tskit_format.generate_schema(
+        variants_chunk_size=variants_chunk_size,
+        samples_chunk_size=samples_chunk_size,
+    )
+    zarr_path = pathlib.Path(zarr_path)
+    vzw = vcz.VcfZarrWriter(TskitFormat, zarr_path)
+    # Rough heuristic to split work up enough to keep utilisation high
+    target_num_partitions = max(1, worker_processes * 4)
+    vzw.init(
+        tskit_format,
+        target_num_partitions=target_num_partitions,
+        schema=schema_instance,
+    )
+    vzw.encode_all_partitions(
+        worker_processes=worker_processes,
+        show_progress=show_progress,
+    )
+    vzw.finalise(show_progress)
+    vzw.create_index()
diff --git a/tests/test_ts.py b/tests/test_ts.py
@@ -0,0 +1,60 @@
+import os
+import tempfile
+
+import numpy as np
+import tskit
+import zarr
+
+from bio2zarr import ts
+
+
+class TestTskit:
+    def test_simple_tree_sequence(self, tmp_path):
+        tables = tskit.TableCollection(sequence_length=100)
+        tables.individuals.add_row(flags=0, location=(0, 0), metadata=b"")
+        tables.individuals.add_row(flags=0, location=(0, 0), metadata=b"")
+        tables.nodes.add_row(flags=tskit.NODE_IS_SAMPLE, time=0, individual=0)
+        tables.nodes.add_row(flags=tskit.NODE_IS_SAMPLE, time=0, individual=0)
+        tables.nodes.add_row(flags=tskit.NODE_IS_SAMPLE, time=0, individual=1)
+        tables.nodes.add_row(flags=tskit.NODE_IS_SAMPLE, time=0, individual=1)
+        tables.nodes.add_row(flags=0, time=1)  # MRCA for 0,1
+        tables.nodes.add_row(flags=0, time=1)  # MRCA for 2,3
+        tables.edges.add_row(left=0, right=100, parent=4, child=0)
+        tables.edges.add_row(left=0, right=100, parent=4, child=1)
+        tables.edges.add_row(left=0, right=100, parent=5, child=2)
+        tables.edges.add_row(left=0, right=100, parent=5, child=3)
+        site_id = tables.sites.add_row(position=10, ancestral_state="A")
+        tables.mutations.add_row(site=site_id, node=4, derived_state="T")
+        site_id = tables.sites.add_row(position=20, ancestral_state="C")
+        tables.mutations.add_row(site=site_id, node=5, derived_state="G")
+        site_id = tables.sites.add_row(position=30, ancestral_state="G")
+        tables.mutations.add_row(site=site_id, node=0, derived_state="A")
+        tables.sort()
+        tree_sequence = tables.tree_sequence()
+        tree_sequence.dump(tmp_path / "test.trees")
+        with tempfile.TemporaryDirectory() as tempdir:
+            zarr_path = os.path.join(tempdir, "test_output.zarr")
+            ts.convert(tmp_path / "test.trees", zarr_path, show_progress=False)
+            zroot = zarr.open(zarr_path, mode="r")
+            assert zroot["variant_position"].shape == (3,)
+            assert list(zroot["variant_position"][:]) == [10, 20, 30]
+
+            alleles = zroot["variant_allele"][:]
+            assert np.array_equal(alleles, [["A", "T"], ["C", "G"], ["G", "A"]])
+
+            genotypes = zroot["call_genotype"][:]
+            assert np.array_equal(
+                genotypes, [[[1, 1], [0, 0]], [[0, 0], [1, 1]], [[1, 0], [0, 0]]]
+            )
+
+            phased = zroot["call_genotype_phased"][:]
+            assert np.all(phased)
+
+            contigs = zroot["contig_id"][:]
+            assert np.array_equal(contigs, ["1"])
+
+            contig = zroot["variant_contig"][:]
+            assert np.array_equal(contig, [0, 0, 0])
+
+            samples = zroot["sample_id"][:]
+            assert np.array_equal(samples, ["tsk_0", "tsk_1"])
