Dataset Card for Dataset Name

The nucleotide_transformer_downstream_tasks dataset features the 18 downstream tasks presented in the Nucleotide Transformer paper. They consist of both binary and multi-class classification tasks that aim at providing a consistent genomics benchmark.

⚠️We note that we have revised and improved our benchmark during the peer-review process. The datasets featured in this repository are used up to this release. We highly encourage to move to the new version available here, which we believe to be much more robust.⚠️

Dataset Description

• Repository: Nucleotide Transformer
• Paper: The Nucleotide Transformer: Building and Evaluating Robust Foundation Models for Human Genomics

Dataset Summary

The different datasets are collected from 4 different genomics papers:

• DeePromoter: Robust Promoter Predictor Using Deep Learning: The datasets features 3,065 TATA promoters and 26,532 non-TATA promoters, with each promoter yielding a negative sequence by randomly sampling parts of the sequence. The promoter_all dataset will feature all the promoters and their negative counterparts, while the promoter_tata and promoter_no_tata respectively provide the TATA and non-TATA parts of the dataset.
• A deep learning framework for enhancer prediction using word embedding and sequence generation: To build the training dataset, the authors collect 742 strong enhancers, 742 weak enhancers and 1484 non-enhancers, and augment the dataset with 6000 synthetic enhancers and 6000 synthetic non-enhancers produced with a generative model. The test dataset is comprised of 100 strong enhancers, 100 weak enhancers and 200 non enhancers. The original paper uses this dataset to do both binary classification (i.e a sample gets classified as non-enhancer or enhancer) and 3-class classification (i.e a sample gets classified as non-enhancer, weak enhancer or strong enhancer). Both tasks are respectively tackled in the enhancers and enhancers_types datasets.
• SpliceFinder: ab initio prediction of splice sites using convolutional neural network: The authors introduce a dataset containing 10,000 samples of donor site, acceptor site, and non-splice-site, resulting in 30,000 total samples that are featured in the splice_sites_all dataset.
• Spliceator: multi-species splice site prediction using convolutional neural networks: Two datasets are introduced by this paper, each of them contain splice sites and their corresponding negative datasets. The dataset splice_sites_acceptor features acceptor splice sites and the other, splice_sites_donor, donor splice sites.
• Qualitatively predicting acetylation and methylation areas in DNA sequences: The paper introduces a set of datasets featuring epigenetic marks identified in the yeast genome, namely acetylation and metylation nucleosome occupancies. Nucleosome occupancy values in these ten datasets were obtained with Chip-Chip experiments and further processed into positive and negative observations to provide the datasets corresponding to the following histone marks: H3, H4, H3K9ac, H3K14ac, H4ac, H3K4me1, H3K4me2, H3K4me3, H3K36me3 and H3K79me3

Dataset Structure

| Task                  | Number of train sequences | Number of test sequences | Number of labels | Sequence length |
| --------------------- | ------------------------- | ------------------------ | ---------------- | --------------- |
| promoter_all          | 53,276                    | 5,920                    | 2                | 300             |
| promoter_tata         | 5,509                     | 621                      | 2                | 300             |
| promoter_no_tata      | 47,767                    | 5,299                    | 2                | 300             |
| enhancers             | 14,968                    | 400                      | 2                | 200             |
| enhancers_types       | 14,968                    | 400                      | 3                | 200             |
| splice_sites_all      | 27,000                    | 3,000                    | 3                | 400             |
| splice_sites_acceptor | 19,961                    | 2,218                    | 2                | 600             |
| splice_sites_donor    | 19,775                    | 2,198                    | 2                | 600             |
| H3                    | 13,468                    | 1,497                    | 2                | 500             |
| H4                    | 13,140                    | 1,461                    | 2                | 500             |
| H3K9ac                | 25,003                    | 2,779                    | 2                | 500             |
| H3K14ac               | 29,743                    | 3,305                    | 2                | 500             |
| H4ac                  | 30,685                    | 3,410                    | 2                | 500             |
| H3K4me1               | 28,509                    | 3,168                    | 2                | 500             |
| H3K4me2               | 27,614                    | 3,069                    | 2                | 500             |
| H3K4me3               | 33,119                    | 3,680                    | 2                | 500             |
| H3K36me3              | 31,392                    | 3,488                    | 2                | 500             |
| H3K79me3              | 25,953                    | 2,884                    | 2                | 500             |