A Novel Computational Framework to Predict Disease-Related Copy Number Variations by Integrating Multiple Data...

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Copy number variations (CNVs) are implicated in complex diseases, but their precise role in cancer pathogenesis is challenging to elucidate due to intricate mechanisms and limited sample sizes.
• The increasing availability of CNV, gene, and disease data presents an opportunity to develop advanced computational frameworks for predicting disease-related CNVs.
• Identifying CNV-disease associations is crucial for understanding disease development and discovering potential therapeutic targets.

Purpose of the Study:

• To develop and validate a novel machine learning framework, IHI-BMLLR, for predicting copy number variation (CNV)-disease path associations.
• To integrate heterogeneous data sources, including CNV, gene, and disease information, to construct a robust biological association network.
• To identify significant CNV-disease pathways and novel candidate genes associated with complex diseases, specifically prostate cancer.

Main Methods:

• Developed IHI-BMLLR (Integrating Heterogeneous Information sources with Biweight Mid-correlation and L1-regularized Logistic Regression under stability selection).
• Utilized self-adaptive biweight mid-correlation (BM) to compute CNV-gene correlations and L1-regularized Logistic Regression (LLR) with stability selection for disease-gene association.
• Employed a weighted path search algorithm to identify top CNV-disease associations within the constructed biological network.

Main Results:

• IHI-BMLLR demonstrated superior performance compared to state-of-the-art methods (CCRET, DPtest) in predicting CNV-disease associations, particularly in controlling false positives.
• Application to prostate cancer data revealed significant path associations, highlighting the method's clinical relevance.
• Identified three novel candidate genes potentially involved in cancer development, warranting further biological investigation.

Conclusions:

• The IHI-BMLLR framework provides a powerful and accurate approach for predicting CNV-disease associations by integrating diverse biological data.
• The method successfully identified novel insights into prostate cancer, including potential new therapeutic targets.
• Future biological validation of the discovered genes is essential to confirm their role in cancer etiology.