Probabilistic inference of biological networks via data integration

Area of Science:

• Systems biology
• Bioinformatics
• Computational biology

Background:

• Understanding subcellular interaction networks is crucial for biological research.
• Supervised network inference uses known interactions to predict new ones.
• Integrating diverse data types enhances the accuracy of functional link prediction.

Purpose of the Study:

• To develop and evaluate a supervised method for inferring gene interaction networks.
• To assess the effectiveness of data integration using multiple kernel learning.
• To improve predictive accuracy through cautious classification and data cleaning.

Main Methods:

• Utilized supervised interactive network inference with a reference set of known links and nonlinks.
• Employed pairwise kernels and multiple kernel learning for data integration.
• Evaluated individual and combined kernel performance, incorporating cautious classification and data cleaning.

Main Results:

• The tensor product pairwise kernel showed strong performance.
• Weighted combinations of different pairwise kernels yielded the highest predictive accuracy.
• Integration of multiple data sources significantly improved network inference accuracy.

Conclusions:

• A data integration approach using multiple pairwise kernels is highly effective for network inference.
• Cautious classification and data cleaning further enhance prediction reliability.
• The developed method achieved high predictive accuracy (up to 99.6%) in yeast (S. cerevisiae).