---
license: mit
tags:
- biology
- plants
- gene expression
pretty_name: Maize and Arabidopsis gene expression
---
# Dataset Card for Maize and Arabidopsis gene expression
Plant Gene expression data used for benchmarking sequence to gene expression prediction ML models. 

## Dataset Details

### Dataset Description

Species included are Maize and Arabidopsis thaliana. Dataset includes gene expression values for leaf and root tissues. 
Within the `tasks` folder, datasets are broken down by `species-task-tissue`, so the structure looks like:
```
    species-task-tissue/
      train.tsv
      validate.tsv
      test.tsv
```
All tasks are split by 80% train, 10% validation, and 10% test. 


- **Curated by:** Taylor Ferebee, Travis Wrightsman, Jingjing Zhai, Aaron Gokaslan, Volodymyr Kuleshov, Edward S. Buckler
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** MIT

### Dataset Sources 
| | | | | | | | | | | | |
|-|-|-|-|-|-|-|-|-|-|-|-|
|sample_name|species|genotype|library_layout|library_selection|reads_location|organ|age|condition|replicate|batch|reference|
|SRR505743|Arabidopsis_thaliana|Col-0|single-read|random|sra|root|seedling|controlled|1|1|SRP013631|
|SRR505744|Arabidopsis_thaliana|Col-0|single-read|random|sra|leaf|seedling|controlled|1|1|SRP013631|
|SRR953400|Arabidopsis_thaliana|Col-0|single-read|random|sra|leaf|seeding|controlled|1|1|PRJNA215448|
|SRR1005386|Arabidopsis_thaliana|Col-0|single-read|random|sra|leaf|seedling|controlled|1|1|PRJNA222364|
|SRR578947|Arabidopsis_thaliana|Col-0|single-read|random|sra|root|seedling|controlled|1|1|SRP013631|
|SRR578948|Arabidopsis_thaliana|Col-0|single-read|random|sra|root|seedling|controlled|1|1|SRP013631|
|ERR2096663|Zea_mays|B73|paired-end|polyA|sra|leaf|seedling|controlled|1|1|PRJEB22166|
|ERR2096664|Zea_mays|B73|paired-end|polyA|sra|leaf|seedling|controlled|1|1|PRJEB22166|
|ERR2096665|Zea_mays|B73|paired-end|polyA|sra|leaf|seedling|controlled|1|1|PRJEB22166|
|ERR2096666|Zea_mays|B73|paired-end|polyA|sra|leaf|seedling|controlled|1|1|PRJEB22166|
|ERR2096667|Zea_mays|B73|paired-end|polyA|sra|leaf|seedling|controlled|1|1|PRJEB22166|
|ERR3773807|Zea_mays|B73|paired-end|polyA|sra|root|seedling|controlled|1|1|PRJEB35943|
|ERR3773808|Zea_mays|B73|paired-end|polyA|sra|root|seedling|controlled|1|1|PRJEB35943|
|ERR986091|Zea_mays|B73|paired-end|random|sra|root|seedling|controlled|1|1|PRJEB10406|



- **Repository:** [https://github.com/maize-genetics/expression-survey]
- **Paper:** PLExBench: A benchmarking suite for predicting gene expression in plants


## Dataset Structure

```
dataset
  genomes/
    Arabidopsis_thaliana/
      annotation.fa
      ath.gff
    Zea_mays/
      annotation.fa
      ath.gff
  tasks/
    species-task-tissue/
      train.tsv
      validate.tsv
      test.tsv
```


### Curation Rationale
 To choose experiments for leaf and root tissues, we focused on datasets that have been used in a recent study and can be found in
multiple databases.

#### Data Collection and Processing
In the max gene expression datasets, for each gene, we take the maximum transcript per million TPM value over experiments. Similarly, for the
absolute expression datasets, we take the mean TPM value over experiments. Finally, for the on-off ex-
pression, we assign 1 to a gene if it has a TPM value in one of the tissues. To create train-test-validation
splits, we use orthogroup guided splitting as introduced by Washburn et al. 2019. Then, we split the training test sets so
that we train on 80% of the orthogroups and test on 10%. Note that for each of the task-based datasets, we
keep the same train-test-validate split. 

**BibTeX:**


## Dataset Card Authors
Taylor Ferebee (tf259@cornell.edu)

## Dataset Card Contact
Taylor Ferebee (tf259@cornell.edu)