TransSE: A Transfer Learning-Based Predictive Model for Distinguishing Super Enhancers and Typical Enhancers.

Super-enhancers (SEs), comprising clusters of transcriptional regulatory elements, play essential roles in gene expression regulation and cell fate determination. Current computational methods for identifying SEs from genomic sequences face challenges in prediction accuracy and cross-species generalizability. This study introduces TransSE, a deep learning framework combining convolutional and recurrent neural networks with cross-species transfer learning for SE prediction. TransSE employs a two-phase strategy: pre-training on combined human and mouse data to learn conserved regulatory features, followed by species-specific fine-tuning to capture organism-specific patterns. Evaluation on human and mouse datasets demonstrates that TransSE achieves superior accuracy compared to existing methods including SENet, DeepSE, and transformer-based models (DNABERT, Enformer), with AUC values of 0.828 (human) and 0.832 (mouse). Ablation studies confirm that convolutional blocks are essential for feature extraction, while transfer learning and recurrent layers provide meaningful performance improvements. Feature analysis reveals that TransSE distinguishes SEs from typical enhancers based on learned sequence representations, achieving 10.4-fold improvement in clustering metrics. The model demonstrates effective cross-species prediction capability, maintaining AUC $\gt $ 0.79 when trained on one species and evaluated on another. Conservation pattern analysis shows that TransSE integrates both evolutionarily conserved and species-specific regulatory features, enabling robust performance across organisms with limited training data. A user-friendly web interface (http://transse.shenlabahmu.com) provides automated SE prediction, motif analysis, and variant impact assessment. Case studies using experimentally validated SEs demonstrate the tool's utility for investigating regulatory elements and prioritizing functional variants in disease-associated genomic regions. TransSE advances the investigation of gene regulatory mechanisms and facilitates identification of disease-associated variation in enhancer regions.