Pathogenic Gene Prediction Algorithm Based on Heterogeneous Information Fusion.

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Complex diseases pose significant health challenges, driving research into identifying causative genes.
• Traditional gene discovery methods are time-consuming and expensive.
• Existing gene prioritization algorithms struggle with sparse gene-disease association data.

Purpose of the Study:

• To develop an advanced algorithm for predicting candidate genes involved in human disease pathogenicity.
• To address the limitations of sparse data and lack of negative evidence in current gene prioritization methods.

Main Methods:

• Proposed the PU induction matrix completion algorithm based on heterogeneous information fusion (PUIMCHIF).
• Utilized compact feature learning to extract gene and disease features from multiple data sources, mitigating data sparsity.
• Implemented a PU-Learning strategy, treating unknown associations as negative examples for biased learning.

Main Results:

• PUIMCHIF demonstrated superior performance in precision, recall, and mean percentile ranking (MPR) compared to existing algorithms.
• Achieved a 50% probability of recovering true gene associations and an MPR of 10.94% in top-100 global prediction analysis.
• Outperformed other methods like IMC and CATAPULT in gene prioritization.

Conclusions:

• The PUIMCHIF algorithm effectively predicts candidate genes for complex diseases by leveraging heterogeneous information fusion and PU-Learning.
• This approach enhances gene prioritization, offering a more efficient and accurate method for identifying disease-related genes.