Argus-NET : A Chromatin-Aware Hybrid CNN-Transformer Framework for the Joint Regression of Multi-Functional Enhancer Activity
Year: 2026
Type: software
Abstract: The Argus-NET architecture integrates the local feature extraction capabilities of Convolutional Neural Networks (CNNs) with the global context-awareness of Transformer blocks. This hybrid approach is specifically engineered to capture the complex, multi-scale syntax of genomic sequences, facilitating the simultaneous prediction of developmental and housekeeping enhancer activities. Repository Contents Model Architecture: Complete source code for the CNN-Transformer hybrid layers. Computation Pipeline: Optimized scripts for data preprocessing, training, and performance benchmarking. Interpretability Modules: Implementation of Integrated Gradients (IG) for motif discovery, systematic ablation studies, and in silico mutagenesis (ISM) analysis. Scientific Rigor and Reproducibility This study corresponds to the code utilized in the manuscript: "Argus-NET: A Chromatin-Aware Hybrid CNN-Transformer Framework for the Joint Prediction of Enhancer Activity." To ensure unbiased evaluation, all experiments were conducted on the GRCh38 (hg38) reference genome using a strict chromosome-held-out validation protocol. By providing this implementation, we aim to uphold the principles of open science and facilitate the reproducible analysis of regulatory genomic elements.
Authors: KARAKUŞ, DOĞANCAN, Batbat, Turgay
Institution: Erciyes University
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7160396923