Identification of Phase-Separation-Protein-Related Function Based on Gene Ontology by Using Machine Learning Methods.

Area of Science:

• Cellular Biology
• Biochemistry
• Computational Biology

Background:

• Phase-separation proteins (PSPs) drive liquid-liquid phase separation, forming cellular compartments.
• Dysregulation of PSPs is implicated in diseases like neurodegenerative disorders and cancer.
• Accurate identification of PSPs is crucial for understanding cellular mechanisms and disease pathology.

Purpose of the Study:

• To identify critical Gene Ontology (GO) terms that define the essential functions of PSPs.
• To develop efficient computational classifiers for identifying PSPs based on their GO terms.
• To enhance the understanding of PSPs' roles in cellular processes and disease.

Main Methods:

• Collected experimentally validated PSPs and non-PSPs as positive and negative samples.
• Extracted Gene Ontology (GO) terms for each protein, creating high-dimensional binary vectors.
• Employed incremental feature selection and integrated feature analysis (including various boosting algorithms and permutation importance) to identify key GO terms and build classifiers.
• Utilized Random Forest (RF) classifiers to distinguish between PSPs and non-PSPs.

Main Results:

• Developed highly accurate Random Forest (RF) classifiers (F1 scores > 0.960) for distinguishing PSPs from non-PSPs.
• Identified significant GO terms crucial for PSP classification, including RNA binding (GO:0003723), membrane organization (GO:0016020), and synaptic function (GO:0045202).
• The computational framework effectively pinpointed classification-relevant GO terms.

Conclusions:

• The study successfully developed efficient RF classifiers for identifying PSPs.
• Key GO terms related to RNA binding, membrane organization, and synaptic function are vital for distinguishing PSPs.
• These findings provide a foundation for future research into the functional roles of PSPs in cellular processes and disease.