SCI-PHY

SCI-PHY performs phylogenomic classification and protein subfamily identification to improve function prediction and detect novel subfamilies.

Key Features:

• Subfamily classification: Automatic identification of protein subfamilies using subfamily Hidden Markov Models (HMMs) constructed from multiple sequence alignments.
• Subfamily HMM construction: Construction of subfamily HMMs to represent conserved sequence patterns within identified subfamilies.
• Efficient scoring scheme: Computationally efficient scoring based on family and subfamily HMMs for high-throughput sequence classification and discrimination of novel subfamilies.
• Logistic regression for novelty detection: Application of logistic regression to distinguish sequences representing novel subfamilies from known ones.
• Information-sharing protocol: Estimation of subfamily HMM parameters using an information-sharing protocol that leverages conservation patterns across related families and subfamilies.
• High specificity and novel subtype prediction: Subfamily HMMs increase separation between homologous and non-homologous proteins, yielding high specificity in classification and enabling prediction of novel protein subtypes.
• Integration with PhyloFacts: Use of pre-calculated subfamily predictions and HMMs from the PhyloFacts resource (Berkeley Phylogenomics Group) covering over 40,000 protein families and domains.

Scientific Applications:

• Functional annotation: Improving accuracy of protein function prediction for genes with limited experimental characterization.
• Evolutionary studies: Supporting phylogenomic analyses and elucidation of evolutionary relationships at the subfamily level.
• Proteome characterization: Enabling detection and classification of novel protein subtypes within proteomic datasets.
• Database search specificity: Enhancing separation of homologous versus non-homologous proteins during sequence database searches.

Methodology:

Uses multiple sequence alignments to construct subfamily HMMs, applies family- and subfamily-based HMM scoring, employs logistic regression for novelty detection, and estimates HMM parameters via an information-sharing protocol, with evaluations measuring separation between homologous and non-homologous proteins.