---
license: cc-by-4.0
---

# SPACE manuscript data

This repository contains the twelve original pre-training split BED
files (human, mouse, ce11 and dm6; train, valid, test) the current Source Data
workbook, and regulatory-probe sample/split manifests (11,000 elements across
two species). For coverage and upstream data access, see the sections below.

BED windows are 131,072 bp. SPACE-Pro constructs 196,608-bp contexts about their
midpoints using the matching reference genome. BED coordinates alone do not
encode the extended sequence. Preserve the genome assemblies and masking used
in the original preprocessing.

Human/mouse processed functional-profile H5 files total approximately 424.85 GB;
prefer their existing hosted dataset rather than uploading a duplicate without
checking the remote contents. Their sequence arrays are 131,072 bp; SPACE-Pro
uses extended genome contexts. Additional-species reference FASTA and BED files
support sequence-only supervision. The local ce11/dm6 functional-profile H5
files are not classified as data used by that self-supervised objective.

Reference assemblies: human hg38 (`hg38.ml.fa` processed file), mouse mm10,
*C. elegans* ce11 and *D. melanogaster* dm6. The exact preparation and source of
the processed human FASTA require documentation before redistribution.

Further downstream dataset split files and access routes must be inventoried
before claiming full reproduction coverage. Dataset-specific redistribution
terms and citations must be preserved. The license below applies only to the
authors' original contributions, not to third-party source material.

Source Data coverage is described in the accompanying notes and workbook index.


## License and attribution

The authors' original manifests, split definitions, documentation and Source
Data contributions are licensed under
[Creative Commons Attribution 4.0 International](LICENSE) (CC BY 4.0).
Attribute these contributions to the authors of **SPACE: A foundation model for
transferable genomic learning across species**, identify the data release used,
provide the license link, and indicate changes when distributing adaptations.

This grant covers only rights held by the SPACE authors. Third-party genome
sequences, annotations, benchmark data, and inherited pre-training split files
retain their original terms; their presence or reference here does not relicense
them. Source citations and any separately identified third-party notices must
be retained. Model weights and software are licensed separately under Apache-2.0.