marin-dna/vertebrate-v1-issue473-center1-cds

marin-dna/vertebrate-v1-issue473-center1-cds

Review status: draft generated for issue #473 review before upload.

Human-anchored 255 bp vertebrate sequences for the cds cohort under the center_1 policy. The policy projects the exact central human nucleotide, requires one target locus, and emits the 255 bp target window centered on that mapped nucleotide. Human anchors come from the fixed #417 protein-coding CDS anchor catalog.

The source projection was produced by the commit-pinned issue #473 workflow at config SHA-256 bf8367c285f955407cfb2dba6102661b2e528261b64fe52095d52a688cd6d039. These publication artifacts were built by the commit-pinned additive publisher.

Splits

• —train: 68,657,166 rows after deterministic reverse-complement augmentation; no chr18 source anchors.
• —validation: 16,384 original-orientation chr18 rows (4,194,304 DNA-character-plus-BOS tokens).

The selected target manifest contains 135 family-deduplicated projection targets. One human-reference row is also emitted per anchor. Train and validation are split by the human source chromosome, not the target locus.

Sequence semantics

All coordinates inside the producer are 0-based, half-open. Every emitted sequence is exactly 255 bases and is oriented to its human anchor. Letter case is preserved from FASTA/2bit inputs: lowercase denotes source repeat masking and uppercase denotes source non-repeat-masked sequence. Conservation scores do not rewrite sequence characters or case.

Center-seeded recovery is not proof that both 127 bp flanks are homologous. Issue #473 reports paired recovery diagnostics and a sampled bidirectional HAL alignment trace separately; users should not substitute target-span length for aligned base coverage.

Schema

• —query_name: String
• —source_chrom: String
• —source_start: Int64
• —source_end: Int64
• —region_label: String
• —species: String
• —alignment_name: String
• —assembly: String
• —taxonomy_id: Int64
• —family: String
• —clade: String
• —phylogenetic_rank: Int64
• —alignment_source: String
• —t_chrom: String
• —t_start: Int64
• —t_end: Int64
• —t_strand: String
• —t_src_size: Int64
• —pre_resize_t_start: Int64
• —pre_resize_t_end: Int64
• —fragment_count: Int64
• —aligned_bases: Int64
• —sequence: String
• —augmentation: String

Intended use and limitations

This dataset is intended for matched-token genomic language-model research. It is not a clinical resource. Assemblies, alignment gaps, repeat masking, family-deduplicated species selection, and the projection acceptance contract all affect the observed sequence distribution. The chromosome-18 split is a pipeline validation split and must not be confused with the protected final variant-effect test chromosomes used by MarinDNA.