Genomic Sequences · living-models
Botanic1-pretraining
A genomic pre-training dataset containing 7.2 million 8,192-base-pair windows across 320 land plant species.
What Botanic1-pretraining is
Botanic1-pretraining is a genomic sequence corpus developed by living-models to support the training of plant genome foundation models. It compiles sequence windows across 320 embryophyte (land plant) species into a unified corpus for machine learning.
The dataset aggregates approximately 59 billion effective base pairs organised into 7.2 million uniform windows of 8,192 base pairs. It provides the core pre-training substrate for the Botanic1 model series, enabling self-supervised sequence representation across plant taxonomy.
How it was made
The corpus is organised into 8,192-base-pair genomic sequence windows stored in JSON text format. The sequence windows are configured for masked token prediction (fill-mask) tasks and can be processed using data pipelines such as Polars, Dask, and Hugging Face Datasets.
Key facts
- Publisher
- living-models
- Total sequence windows
- 7.2 million
- Window size
- 8,192 bp
- Effective volume
- Approximately 59 billion base pairs
- Taxonomic scope
- 320 embryophyte species
- Task category
- fill-mask
- Format
- JSON
What you can use it for
- Plant Foundation Model Pre-training
- Serving as the base sequence corpus to pre-train genomic language models across diverse plant lineages.
- Masked Sequence Modelling
- Executing fill-mask self-supervised learning objectives on 8,192-base-pair genomic windows.
- Cross-Species Sequence Representation
- Benchmarking and learning nucleotide representations across 320 distinct land plant genomes.
Explore the data
This dataset is gated
Its authors require their terms to be accepted before the data can be read. Accept them on the original dataset page. The platform does not accept terms on your behalf, so the rows are not shown here.Strengths and limits
Good at
- Covers 320 embryophyte species, providing broad representation across land plant genomics.
- Contains 59 billion effective base pairs across 7.2 million standardized 8,192-base-pair windows.
- Compatible with standard distributed data processing tools including Dask, Polars, and Hugging Face Datasets.
Watch out for
- Coverage is restricted to embryophyte species, omitting non-plant taxa and non-embryophyte algae.
- Genomic data is segmented into fixed 8,192-base-pair windows, which may not fit workflows requiring variable-length inputs.
Questions
Can I look inside Botanic1-pretraining without downloading it?
Not here: its authors gate it behind terms each user must accept on the original page. Its description and statistics are still shown.
What licence is Botanic1-pretraining under?
Its authors declare other. Read the terms on the original page before reusing the data.
Which organisms are represented in Botanic1-pretraining?
The dataset contains sequences spanning 320 embryophyte (land plant) species.
What is the length of each sequence window?
Every training window in the dataset has a fixed length of 8,192 base pairs.
What machine learning task was this dataset designed for?
It is configured for fill-mask (masked sequence modelling) pre-training tasks.
Written from Botanic1-pretraining's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.