NIAPU

NIAPU applies positive-unlabeled learning to prioritize and identify candidate disease genes from biological networks using Markov diffusion-based multi-class labeling and derived network features.

Key Features:

• Positive-Unlabeled Learning Framework: Uses positive-unlabeled learning to prioritize candidate disease genes when only a subset of positive instances is annotated.
• Network-Based Features: Computes novel network-based features derived from Markov diffusion to capture complex interactions within biological networks.
• Multi-Class Labeling Strategy: Implements a Markov diffusion-based multi-class labeling algorithm to assign labels across multiple classes for gene prioritization.
• Comparative Analysis: Compares NIAPU-derived features against classical topological, functional/ontological, and other network- and biology-derived features.
• Evaluation Across Datasets and Algorithms: Evaluates performance on ten disease datasets using three distinct machine learning algorithms to validate predictive power.

Scientific Applications:

• Gene-Disease Association Prioritization: Prioritizes candidate gene-disease associations that are unannotated or under-studied.
• Candidate Disease Gene Discovery: Identifies putative disease genes for downstream experimental validation.
• Benchmarking of Network Features: Facilitates comparative assessment of network- and biology-derived features against state-of-the-art algorithms for gene discovery.

Methodology:

Positive-unlabeled learning; Markov diffusion-based multi-class labeling to derive network-based features; comparative analyses against topological, functional/ontological and other network- and biology-derived features; evaluation on ten disease datasets using three machine learning algorithms.