ProMK

ProMK integrates heterogeneous proteomic data by transforming them into kernels and jointly optimizing a composite kernel and multi-label classifiers to improve protein function prediction.

Key Features:

• Kernel Transformation: Transforms various proteomic data sources into kernels to represent heterogeneous information for integration.
• Composite Kernel Integration: Integrates individual kernels into a composite kernel for combined analysis.
• Simultaneous Optimization: Iteratively optimizes both the weights of the kernels and the empirical loss of multi-label classifiers concurrently.
• Selective Kernel Weighting: Selectively integrates kernels and downgrades the influence of noisy kernels to minimize the impact of unreliable data sources.
• Label-wise Weight Optimization with Correlations: Optimizes kernel weights for each label concurrently while considering correlations among labels to address computational challenges in multi-class or multi-label data.

Scientific Applications:

• Functional Annotation: Assigns functions to proteins based on integrated proteomic kernels.
• Biomarker Discovery: Identifies candidate biomarkers by predicting protein functions linked to disease states.
• Drug Target Identification: Facilitates discovery of potential drug targets through improved functional prediction of proteins.

Methodology:

Kernel transformation: converting heterogeneous proteomic data into kernels.
Composite kernel formation: integrating these kernels into a composite form.
Iterative optimization: simultaneously optimizing kernel weights and reducing empirical loss for each label in a multi-label classification setting.