Rice Drought-Response Gene Variant Effect Prediction Pipeline

Overview

This repository contains a complete computational pipeline for predicting the functional impact of genetic variants in rice (Oryza sativa) genes using DeepChem-variant

Repository Contents

Files Included

1. deepvariant_case_study_rice_osdreb2a.ipynb

   • Complete Jupyter notebook with full pipeline implementation
   • Runnable in Google Colab (GPU recommended)

2. best_model (1).pth

   • Pre-trained DeepChem-Variant model weights
   • Trained on indica rice subspecies variants
   • 95.3% accuracy on validation set
   • Can be used for transfer learning to other rice varieties

Required Data Files

Download the following files from RAP-DB (Rice Annotation Project Database):

Download URL: https://rapdb.dna.affrc.go.jp/download/irgsp1.html

Required files:

1. Reference Genome (FASTA format)

   • File: IRGSP-1.0_genome.fasta.gz
   • Description: Complete rice reference genome sequence

2. Gene Annotations (GTF/GFF format)

   • File: IRGSP-1.0_representative_annotation_*.gtf.gz
   • Description: Gene structure and coding sequence annotations

3. CDS Sequences (FASTA format)

   • File: IRGSP-1.0_cds_*.fasta.gz
   • Description: Coding sequences for all annotated genes

4. Protein Sequences (FASTA format)

   • File: IRGSP-1.0_protein_*.fasta.gz
   • Description: Translated protein sequences

Model Performance

Dataset	Accuracy	Precision	Recall	F1-Score
Training/Validation (indica)	95.37%	95.29%	95.37%	95.32%
Test Set (same subspecies)	95.30%	95.39%	95.30%	95.32%
Test Set (different subspecies)	95.94%	96.06%	95.94%	95.98%