iPromoter-Seqvec

iPromoter-Seqvec predicts promoter regions in human and mouse genomes, distinguishing TATA and non-TATA promoters to support analysis of transcriptional regulation and gene structure.

Key Features:

• BiLSTM Neural Networks: Leverages bidirectional long short-term memory (BiLSTM) networks to capture sequential dependencies within DNA sequences.
• Sequence-Embedded Features: Extracts sequence-embedded features from input DNA sequences to provide rich representations for promoter classification.
• Benchmark Datasets: Uses promoter and non-promoter sequences sourced from the Eukaryotic Promoter database and refined into four benchmark datasets for training and validation.
• Performance Metrics: Evaluates predictive performance using area under the receiver operating characteristic curve (AUCROC) and area under the precision-recall curve (AUCPR), reporting AUCROC values of 0.85–0.99 and AUCPR values of 0.86–0.99 and outperforming other state-of-the-art methods.
• Artificial Non-Promoter Sequences: Constructs artificial non-promoter sequences based on promoter sequences to enable learning of specific discriminative characteristics.
• Robustness and Stability: Demonstrates stable prediction of both TATA and non-TATA promoters across human and mouse genomes, with slightly higher predictive power for TATA promoters.

Scientific Applications:

• Promoter identification and annotation: Locates and annotates promoter regions in human and mouse genomic sequences.
• Transcriptional regulation studies: Supports investigation of transcription initiation and regulatory mechanisms by mapping promoter locations.
• Gene structure discovery: Aids in delineating gene structures by identifying upstream promoter elements associated with transcription start sites.

Methodology:

The method applies BiLSTM networks to sequence-embedded features, is trained on four benchmark datasets derived from the Eukaryotic Promoter database using artificially constructed non-promoter sequences, and is evaluated with AUCROC and AUCPR metrics.