afg1/plantrnafm_preprocessed_transcriptome

PlantRNAFM preprocessed transcriptome

This dataset was not prepared by the authors of the original paper, it is my attempt to reproduce their published preprocessing pipeline

PlantRNAFM is a foundation model for plant RNA, published in Nature Machine Intelligence: https://www.nature.com/articles/s42256-024-00946-z

The dataset on which it was trained contains, in part, a set of processed transcriptomes where the 5', CDS and 3' regions are labelled. One of the tasks on which the foundation model was trained is to label these regions.

The original dataset comes from the onekp project, where the full transcriptome of over 1000 plants is available. This dataset is a preprocessed version of that data, packaged in parquet.

The preprocessing

Transcriptomes and proteomes for >1400 plants were downloaded and we run the following pipeline:

1. Use tblastn to map the translated proteins onto the transcribed nucleotides
2. Filter for best hit with evalue 1e-10 and qcov_hsp_perc 70.0
3. Convert hits into parquet, along with a representation of the 5'/CDS/3' labelling

The preprocesing pipeline is available here: https://github.com/afg1/plantrnafm_preprocessing

What do I do with it

This dataset should be equivalent to the one used for the region labelling task in the PlantRNAFM pretraining task, so you could use it for tasks like that.