Optimisation Models for Pathway Activity Inference in Cancer.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• High-throughput technologies generate vast, complex molecular data for disease profiling.
• Accurate classification of disease phenotypes from this data is challenging.
• Biological pathways offer a way to incorporate prior knowledge, reduce dimensionality, and improve interpretability of machine learning models.

Purpose of the Study:

• To develop a novel mathematical optimization model for pathway activity inference.
• To achieve precise disease phenotype prediction using gene expression data.
• To enhance the interpretability of machine learning models in biological data analysis.

Main Methods:

• A mixed-integer linear programming (MILP) optimization model was developed.
• Pathway activity is inferred as a linear combination of gene expression, weighted by optimized gene importance.
• The model maximizes discrimination between phenotype classes and minimizes misallocation of samples.

Main Results:

• The model was evaluated on breast and colorectal cancer transcriptome data.
• It demonstrated good optimality and prediction performance for cancer subtypes.
• Comparative analysis showed competitive prediction accuracy, robustness, and interpretability against other methods.

Conclusions:

• Mathematical programming provides an efficient computational strategy for pathway activity inference.
• The proposed model offers a flexible and interpretable approach for disease subtype prediction.
• This method yields insights into key pathways and genes driving disease phenotypes.