deepTFBS: Improving within- and Cross-Species Prediction of Transcription Factor Binding Using Deep Multi-Task and...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of transcription factor binding sites (TFBSs) is essential for understanding gene regulation.
• Existing methods face challenges in predicting TFBSs, especially across different plant species or with limited experimental data.

Purpose of the Study:

• To present deepTFBS, a deep learning (DL) framework for robust TFBS prediction within and across plant species.
• To leverage multi-task DL and transfer learning for enhanced TFBS prediction accuracy, particularly in low-data scenarios.

Main Methods:

• Developed deepTFBS, a comprehensive DL framework utilizing a DNA language model to capture TF binding grammar.
• Employed multi-task DL and transfer learning to utilize large-scale TF binding profiles for improved predictions.
• Evaluated performance against established methods like position weight matrix, deepSEA, and DanQ.

Main Results:

• deepTFBS significantly outperformed baseline models in predicting TFBSs in Arabidopsis, showing substantial improvements in the area under the precision-recall curve (PRAUC).
• Demonstrated superior cross-species TFBS prediction accuracy in wheat compared to existing strategies, with a notable PRAUC increase.
• Showcased the utility of incorporating gene conservation and binding motif information for efficient TFBS prediction.

Conclusions:

• deepTFBS offers a powerful and versatile tool for TFBS prediction in plants, enhancing our understanding of gene regulation.
• The framework's ability to perform cross-species predictions and utilize limited data makes it valuable for diverse genomic studies.
• A case study with the WUSCHEL transcription factor highlights deepTFBS's potential for practical applications in comparative genomics.