iPT-BFL: A TATA promoter prediction model based on Boundary Focus Loss

The promoter, as a short DNA sequence adjacent to the start codon, plays a crucial role in transcriptional regulation in biology. Therefore, accurate identification of DNA promoters is of great practical significance for a deeper understanding and manipulation of biological transcription processes. However, most existing DNA promoter recognition methods only focus on the recognition of promoter strength or sigma type, ignoring the crucial impact of TATA promoter on transcription processes. Given this limitation, it is particularly important to accurately determine whether the promoter is the TATA promoter. To achieve the preparation recognition of the TATA promoter, this paper innovatively proposes a model called iPT-BFL based on deep learning technology, specifically designed for identifying the TATA promoter. The iPT-BFL model integrates advanced architectures such as embedding layers, bidirectional long short-term memory neural networks, and convolutional layers, and introduces the Boundary Focal Loss (BFL) as the loss function to achieve accurate prediction of the TATA promoter. After comprehensive experimental verification, the iPT-BFL model has demonstrated excellent TATA promoter prediction ability. Compared with the latest TATA promoter prediction tool iPTT-CNN, iPT-BFL has achieved significant improvements in various evaluation indicators.