NetGO 2.0: improving large-scale protein function prediction with massive sequence, text, domain, family and network...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Automated protein function prediction (AFP) faces challenges due to the rapid increase in protein sequence data.
• AFP is treated as a large-scale multi-label classification problem, with proteins associated with numerous Gene Ontology (GO) terms.
• The previous NetGO tool utilized a learning to rank (LTR) framework with network and sequence information for effective AFP.

Purpose of the Study:

• To introduce NetGO 2.0, an advanced version of the NetGO tool for large-scale automated protein function prediction.
• To enhance AFP performance by incorporating literature information via logistic regression and deep sequence information via recurrent neural networks (RNNs).
• To evaluate the effectiveness of NetGO 2.0 using datasets generated according to the Critical Assessment of Functional Annotation (CAFA) protocol.

Main Methods:

• NetGO 2.0 integrates literature data using logistic regression and deep sequence features using RNNs within the LTR framework.
• The study generated datasets adhering to the CAFA protocol for rigorous evaluation.
• Performance was compared against the previous NetGO tool.

Main Results:

• NetGO 2.0 demonstrated significant performance improvements over NetGO, particularly in Biological Process Ontology (BPO) and Cellular Component Ontology (CCO).
• A 12.6% increase in Area Under the Precision-Recall Curve (AUPR) for BPO was observed.
• A 2.6% improvement in Fmax for CCO was achieved.

Conclusions:

• The integration of text (literature) and deep sequence information substantially benefits the functional annotation of proteins, especially for BPO and CCO.
• NetGO 2.0 represents a significant advancement in large-scale automated protein function prediction.
• The NetGO 2.0 web server is publicly accessible for research use.