ProkBERT Promoter Evaluation

The model presented below is a deep learning classifier based on ProkBERT, trained on large genomic datasets to evaluate DNA sequences directly from the raw nucleotide sequence. It analyses k-mer representations in context to detect promoter-like patterns, including subtle and non-canonical motifs that feature-based models may miss.

This approach provides high sensitivity and the ability to detect complex promoter signals, capturing regulatory patterns beyond predefined features. However, it is less interpretable, more computationally intensive, and confidence values can vary slightly between runs due to its deep learning architecture.