FunCProp

FunCProp applies parameterizable Bayesian priors and machine-learning integration of genome-wide datasets to improve protein function prediction by statistically informing the selection of negative examples for Gene Ontology annotations.

Key Features:

• Negative example selection: Implements a novel approach to selecting negative examples (proteins that do not possess a given function) for training predictive models.
• Parameterizable Bayesian priors: Derives adjustable Bayesian priors from comprehensive annotation data to inform classification.
• Priors used in prediction: Incorporates the Bayesian priors directly during the prediction process to influence model outputs.
• Integration with GeneMANIA: Integrates FunCProp methodology into the GeneMANIA algorithm for network-based function prediction.
• Multi-omic data integration: Integrates multiple genome-wide data types using advanced machine learning techniques.
• Benchmark performance: Demonstrates improved accuracy across various metrics when tested on yeast and mouse proteomes.
• Gene Ontology focus: Targets prediction of protein functions using Gene Ontology annotations.

Scientific Applications:

• Protein function prediction: Enhances prediction of protein functions annotated with Gene Ontology terms.
• Supervised model training: Provides statistically informed negative sets for training supervised classifiers of protein function.
• Network-based inference: Augments network-based algorithms such as GeneMANIA for improved functional inference.
• Cross-species benchmarking: Serves as a method for evaluating and improving predictive performance on yeast and mouse proteomes.

Methodology:

Derives parameterizable Bayesian priors from annotation data, uses those priors to select negative examples and to inform predictions, and integrates these components into the GeneMANIA algorithm while combining multiple genome-wide data types with advanced machine learning techniques.