DeepPheno

DeepPheno predicts gene-phenotype associations from gene loss-of-function mutations by leveraging Gene Ontology (GO) functional annotations to infer Human Phenotype Ontology (HPO) classes.

Key Features:

• Neural network classifier: Uses a neural network-based hierarchical multi-class multi-label classification approach for phenotype prediction.
• Two-step prediction: First predicts functional annotations (GO classes) for gene products and then infers HPO phenotypic outcomes for single-gene loss-of-function mutations.
• Ontology-based hierarchical classification: Implements an ontology-aware classifier tailored for large-scale hierarchical classification consistent with GO and HPO structures.
• Genome-wide coverage: Produces comprehensive predictions across all known protein-coding genes.
• Evaluation with CAFA metrics: Performance assessed using Critical Assessment of Functional Annotation (CAFA) challenge metrics and compared with top-performing CAFA2 methods and other state-of-the-art approaches.
• Database integration: Predictions have been incorporated into phenotype databases to extend known genotype-phenotype associations.

Scientific Applications:

• Forward genetic screens: Supports mapping from genotype perturbations to predicted phenotypes for experimental target selection.
• Reverse genetic screens: Aids identification of genes likely to produce observed phenotypes when perturbed.
• Gene-disease association studies: Facilitates comparison of predicted HPO phenotypes with disease phenotypes to identify candidate disease genes.
• Expansion of genotype-phenotype repositories: Contributes predicted associations to phenotype databases to broaden resources for genetic research.

Methodology:

Employs a neural network-based hierarchical multi-class multi-label classification framework with a two-step procedure that predicts GO functional annotations for gene products and subsequently infers HPO phenotypes using an ontology-based classifier.