Analysis of protein function and its prediction from amino acid sequence

Area of Science:

• Genomics
• Proteomics
• Bioinformatics

Background:

• Understanding protein function is vital for molecular biology and human disease research.
• Millions of uncharacterized proteins from genomic sequencing necessitate automated function prediction.
• Current methods for protein functional annotation are limited by throughput and accuracy.

Purpose of the Study:

• To estimate the accuracy of transferring Gene Ontology (GO) terms from protein sequences.
• To develop and evaluate a novel computational predictor for protein function.
• To improve the prediction of protein function using amino acid sequence data.

Main Methods:

• Analysis of GO term transfer accuracy using pairwise sequence alignments.
• Development of a multioutput neural network framework for functional annotation (FANN).
• Evaluation of FANN-GO against standard sequence identity (SID) methods and GOtcha.

Main Results:

• Transferring GO terms via sequence alignment shows only moderate accuracy.
• Sequence identity (SID) has a surprisingly limited influence on GO term transfer accuracy across a wide range (30-100%).
• The developed FANN-GO predictor outperforms existing methods like global/local SID transfer and GOtcha.

Conclusions:

• Automated protein function prediction is essential due to the vast number of sequenced proteins.
• FANN-GO offers a more accurate approach to predicting protein function from amino acid sequences.
• The neural network framework effectively models dependencies between functional terms for improved annotation.