language: en
license: apache-2.0
library_name: pytorch
pipeline_tag: text-classification
tags:
- genomics
- mutation
- pathogenicity
- splice
- explainable-ai
- biology
- clinical-ai
🧬 MutationPredictorCNN_v2 — Splice-Aware Pathogenicity Predictor
Model Summary
MutationPredictorCNN_v2 is a splice-aware convolutional neural network designed to predict pathogenicity of single nucleotide variants using genomic sequence context and splice-aware features.
Supports built-in explainability:
• CNN activation heatmap
• Gradient attribution
• Counterfactual mutation analysis
• Feature ablation analysis
• Splice distance analysis
Validation accuracy: 74.8%
Intended Use
Research use cases:
• Genomic variant interpretation
• Explainable AI research
• Variant prioritization
• Educational and academic research
NOT intended for clinical diagnostic use.
Model Architecture
CNN-based architecture:
Input: 1106 features
Output: Pathogenicity probability
Explainability heads:
• Mutation importance
• Region importance
• Splice importance
Training Data
Source: ClinVar
Dataset size:
100,000 variants
50,000 pathogenic
50,000 benign
Sequence window: 99 bp
Performance
Validation accuracy:
74.8%
Balanced dataset.
Explainability
Provides multi-level explainability:
• Activation heatmap
• Mutation rank percentile
• Gradient attribution map
• Counterfactual analysis
• Feature ablation analysis
Limitations
Supports only:
• Single nucleotide variants
• 99 bp context window
Does not include:
• Conservation scores
• Protein structure
• Expression context
Disclaimer
⚠ Research use only
Not a clinical diagnostic tool
Maintainer
Nilesh Hanotia