Integration of relational and hierarchical network information for protein function prediction

Abstract

Background: In the current climate of high-throughput computational biology, the inference of a protein's function from related measurements, such as protein-protein interaction relations, has become a canonical task. Most existing technologies pursue this task as a classification problem, on a term-by-term basis, for each term in a database, such as the Gene Ontology (GO) database, a popular rigorous vocabulary for biological functions. However, ontology structures are essentially hierarchies, with certain top to bottom annotation rules which protein function predictions should in principle follow. Currently, the most common approach to imposing these hierarchical constraints on network-based classifiers is through the use of transitive closure to predictions.

Combining Hierarchical Inference in Ontologies with Heterogeneous Data Sources Improves Gene Function Prediction

“…In order to obtain a sense of the overall performance gains offered by the various components of our proposed method, we compared it to two other related methods proposed recently in the literature: the hierarchical Binomialneighborhood (HBN) method [6] and the heterogeneous Binomial-neighborhood (HeteroBN) method [11]. Each of these methods was referred to earlier and differs from our PHIPA method in important aspects of integration.…”

“…Employing the Hierarchical Binomial-Neighborhood (HBN) assumption from [6], we can show that for a given network (omitting network index ), is labeled (NOT labeled) with @ . We estimate them from the training data using a standard pseudo-likelihood approach.…”

“…For biological process and pathway annotation, the use of protein interaction relationships in terms of functional linkage graphs has been a popular choice in recent years. Markov Random Field model are well suited to model such relationships, and are commonly applied under a Bayesian framework [6,9,11]. Many such relationships and annotations are stored in various databases, such as BIOGRID, a protein-protein interaction (PPI) dataset; MIPS, a database for genome/protein sequences, and the Gene Ontology (GO) database, a rigorous vocabulary for biological functions and available for computation.…”

“…However, importantly, all of these that predict at multiple depths in the GO hierarchy take a separate step to correct inconsistent predictions, rather than producing them directly in a probabilistically coherent way. This problem is tackled in [6], but the methodology proposed therein is limited in that it uses only a simple PPI network as input.…”

Micro-Array-Based Analysis of Cell-Cycle Gene ExpressionDuring Spermatogenesis in the Mouse.

“…Expression data were normalized within GeneSpring using normalization methods as previously described by Shima et al [21]. To identify cell-cycle genes present on the chip, we examined the controlled vocabulary for the description of cellular components, molecular functions, and biological processes provided by Gene Ontology (GO) using the primary annotation term GO:0007049 and its subheadings as described previously [28,29]. Of 45 101 probe sets present on the chip, 1923 were annotated by GO to be involved in cell-cycle control (Fig.…”