Learning with multiple pairwise kernels for drug bioactivity prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Bioinformatics inference problems, such as drug bioactivity prediction, are often modeled as pairwise learning tasks.
• Multiple Kernel Learning (MKL) integrates diverse biomedical data but faces computational challenges due to large pairwise kernel spaces.

Purpose of the Study:

• To introduce pairwiseMKL, a novel method for efficient learning with multiple pairwise kernels.
• To address the computational bottlenecks in existing MKL algorithms for large-scale pairwise learning problems.

Main Methods:

• pairwiseMKL efficiently learns mixture weights and prediction functions without explicit computation of massive pairwise matrices.
• The method is designed for time- and memory-efficient processing of large datasets.

Main Results:

• pairwiseMKL demonstrates high performance in quantitative drug bioactivity prediction tasks, including anticancer efficacy and target profiling.
• The method achieves accurate predictions with sparse solutions, automatically identifying relevant data sources.

Conclusions:

• pairwiseMKL offers a computationally feasible solution for large-scale pairwise learning problems in bioinformatics.
• The approach facilitates accurate drug bioactivity prediction and highlights the importance of integrating multiple data sources.