Using Fully Convolutional Network to Locate Transcription Factor Binding Sites Based on DNA Sequence and Conservation...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcription factors (TFs) regulate gene expression by binding to DNA.
• Identifying TF binding sites (TFBSs) is crucial for understanding gene regulation.
• Deep learning (DL) methods have shown promise in predicting TFBSs, but often focus on sequence presence rather than precise location.

Purpose of the Study:

• To develop a novel deep learning model for precise nucleotide-level TFBS prediction.
• To improve the accuracy and resolution of TFBS identification compared to existing methods.

Main Methods:

• A fully convolutional network (FCN) was developed, incorporating refinement residual blocks (RRB) and a global average pooling layer (GAPL).
• The FCNARRB model was designed to classify binding sequences at the nucleotide level, outputting dense labels.
• Human ChIP-seq datasets were utilized for model training and evaluation.

Main Results:

• The FCNARRB model demonstrated significant improvements in performance.
• Global Average Pooling Layer (GAPL) addition boosted performance by 9.32% (IoU) and 7.61% (PRAUC).
• Refinement Residual Block (RRB) addition improved performance by 7.40% (IoU) and 4.64% (PRAUC).
• Conservation information was identified as a beneficial factor for TFBS localization.

Conclusions:

• The proposed FCNARRB model effectively predicts TFBSs at the nucleotide level.
• The RRB and GAPL components significantly enhance model performance for TFBS identification.
• Integrating conservation information can further improve the accuracy of TFBS prediction.