Predicting protein phosphorylation sites in soybean using interpretable deep tabular learning network.

Area of Science:

• Plant molecular biology
• Biochemistry
• Computational biology

Background:

• Protein phosphorylation is a critical post-translational modification regulating plant functions like signaling and gene expression.
• Dysregulated phosphorylation is linked to plant diseases, making accurate prediction of phosphorylation sites (p-sites) essential.
• Experimental methods for p-site prediction are costly and error-prone.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting protein phosphorylation sites in soybean.
• To introduce a high-performance, interpretable deep tabular learning network (TabNet) for this task.
• To utilize a hybrid feature set for predicting serine, threonine, and tyrosine p-sites.

Main Methods:

• Collected experimentally verified soybean p-sites data from public databases.
• Employed a hybrid feature set including sequential, physicochemical, and PSSM features.
• Developed and compared machine learning models, focusing on TabNet with a window size of 13.
• Removed redundant samples by filtering protein sequences with >40% similarity.

Main Results:

• The developed machine learning techniques achieved over 70% accuracy.
• TabNet, using the hybrid feature set and window size 13, demonstrated superior performance.
• TabNet achieved 78.96% sensitivity and 77.24% specificity.
• The model showed high performance and interpretability, outperforming traditional methods.

Conclusions:

• TabNet offers an effective, automated, and interpretable method for predicting plant protein p-sites.
• This approach can aid in understanding plant signaling pathways and identifying disease-related phosphorylation events.
• The developed technique provides a valuable tool for plant research, reducing experimental costs and errors.