BERT-Promoter

BERT-Promoter predicts DNA promoters and classifies promoter strength to support studies of transcriptional regulation and gene expression.

Key Features:

• Transformer-Based Encoding: DNA sequences are encoded using a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model to capture complex sequence patterns and dependencies.
• Explainable Machine Learning (SHAP): SHapley Additive exPlanations (SHAP) analysis is applied to BERT-derived features to identify and quantify the contribution of sequence elements to predictions.
• Machine Learning Classifiers: Classifiers are trained on top-ranked features to distinguish promoters from non-promoters and to classify promoter activity as strong or weak.

Scientific Applications:

• Gene regulation studies: Prediction and classification of promoters provide insights into transcription initiation and regulatory element function.
• Disease-associated promoter analysis: Characterization of promoter activity supports investigation of links between aberrant promoter function and disease.
• Genomics and transcriptomics: Promoter identification contributes to genome annotation and interpretation of transcriptional datasets.
• Personalized medicine: Promoter activity classification can inform studies of individual regulatory variation relevant to precision medicine.

Methodology:

DNA sequences are encoded with a pre-trained BERT model; SHAP analysis selects top-ranked features from the BERT encodings; machine learning classifiers use these features to predict promoter presence and to classify promoter strength (strong vs weak) and distinguish non-promoters.