GeneRAIN Dataset: Training Data, Gene Embeddings, and Model Checkpoints

This dataset contains various files essential for understanding and employing the GeneRAIN models, as described in the accompanying manuscript. GeneRAIN models use bulk RNA-seq data and a 'Binning-By-Gene' normalization method. These models aim to improve upon existing methods in understanding biological information and include a vector representation of genes called GeneRAIN-vec. After thorough testing, these models have shown their effectiveness in predicting a wide range of biological characteristics, including for long non-coding RNAs. This shows their usefulness and potential in bioinformatics and computational biology. The provided dataset includes:

Gene Embedding Files: These files offer 200-dimensional and 32-dimensional vector representations of genes.

Checkpoint Files:  Checkpoints of various GeneRAIN models.

JSON Mapping Files: For gene to index mapping and tokenization processes. Note that some genes with low mean expression values might not be present in the model input dataset.

ARCHS Human Bulk RNA-seq Data: Access the 'human_gene_v2.2.h5' file and corresponding metadata from the official ARCHS4 website.

Normalized ARCHS Dataset: Processed via the 'Binning-By-Gene' method, this dataset is divided into five sample-based segments, ready for model training.

Mean Expression and Flag Files: Contains mean expression values of genes and boolean flags to help filter duplicate gene symbols.

Normalization and Binning Files: Utilize these with the 'normalize_expr_mat.ipynb' notebook to determine binning boundaries in new expression data.

Example Input/Output: Provided for 'anal_dataset.ipynb' to demonstrate model application.

Prediction Results: The 'genes_clf_pred_results.parquet' file contains the predicted results of coding and lncRNA genes.