GraphPro: An interpretable graph neural network-based model for identifying promoters in multiple species.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate promoter identification is vital for understanding gene expression and disease mechanisms.
• Experimental methods for promoter identification are costly and time-consuming.
• Developing efficient computational models for promoter identification is essential.

Purpose of the Study:

• To introduce GraphPro, a novel interpretable graph neural network model for multi-species promoter identification.
• To enhance the accuracy and cross-species prediction ability of computational promoter identification.
• To improve the biological interpretability of promoter identification models.

Main Methods:

• Encoding DNA sequences using k-tuple frequency, physicochemical properties, and dna2vec.
• Utilizing convolutional neural networks and graph neural networks for feature extraction.
• Employing a fully connected neural network for promoter prediction.
• Validating the model on eight species datasets, including Human, Mouse, and E. coli.

Main Results:

• GraphPro achieved average Sn, Sp, Acc, and MCC values of 0.9123, 0.9482, 0.8840, and 0.7984, respectively.
• Demonstrated superior recognition accuracy across multiple species compared to previous methods.
• Outperformed existing methods in cross-species prediction capabilities.
• Validated biological interpretability through visualization and analysis of transcription factor binding motifs.

Conclusions:

• GraphPro offers a powerful and interpretable tool for accurate multi-species promoter identification.
• The model advances computational approaches in genomics and gene regulation studies.
• GraphPro's interpretability provides valuable insights into transcription factor binding and promoter function.