# GraphTransNet: predicting epilepsy-related genes using a graph-augmented protein language model

## TL;DR

GraphTransNet is a new AI model that accurately predicts genes linked to epilepsy, helping improve diagnosis and drug development.

## Contribution

GraphTransNet introduces a hybrid neural network using graph-augmented protein language models for predicting epilepsy-related genes.

## Key findings

- GraphTransNet outperforms existing tools in precision and recall for identifying epilepsy-related genes.
- The model integrates transformer and CNN components for improved prediction accuracy.
- Results show GraphTransNet's effectiveness compared to traditional machine learning and deep learning methods.

## Abstract

Introduction: Epilepsy, a complex neurological disorder characterised by recurrent seizures and significant genetic heterogeneity, presents considerable challenges form accurate diagnosis and drug target identification. While traditional genomewide association studies (GWAS) and sequencing technologies have advanced our understanding of epilepsy-related gene targets, they often struggle to identify novel and rare variants crucial for precise diagnosis and targeted drug development. The increasing availability of large-scale genomic data, coupled with the power of deep learning, offers a promising avenue for progress.

Method: In this work, we introduce GraphTransNet, a novel hybrid neural network model designed for predicting epilepsy-associated gene targets, with direct implications for improved disease diagnosis and therapeutic target identification. GraphTransNet leverages protein language models (specifically ESM) to generate numerical embeddings from gene sequences. These embeddings are then processed by a novel architecture integrating transformer and convolutional neural network (CNN)components to predict epilepsy-related gene targets.

Results: Our results demonstrate that GraphTransNet achieves high accuracy in identifying epilepsy targets, outperforming existing predictive tools in terms of both recall and precision metrics for reliable disease diagnosis and effective drug target identification. Rigorous comparisons with established machine learning methods and other deep learning architectures further underscore the efficacy of GraphTransNet.

Discussion: This approach represents a valuable computational tool for advancing epilepsy genetics research, with the potential to contribute to more accurate diagnostic strategies and the discovery of novel drug targets for improved treatment outcomes.