HABiC: an algorithm based on the exact computation of the Kantorovich-Rubinstein optimizer for binary classification...

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Oncology

Background:

• Machine learning (ML) in oncology relies on omics data for precision medicine.
• High-dimensional omics data present mathematical challenges like multicollinearity for ML algorithms.
• Novel algorithms are needed to overcome these limitations for clinical application.

Purpose of the Study:

• To develop a robust prediction algorithm for omics data analysis in oncology.
• To enhance the precision and accuracy of ML models in clinical settings.
• To address challenges posed by high dimensionality and multicollinearity in omics datasets.

Main Methods:

• Developed a prediction algorithm utilizing the 1-Wasserstein distance for complex variable relationships.
• Employed exact computation of the Kantorovich-Rubinstein optimizer for increased algorithm precision.
• Incorporated dimension reduction and aggregation methods to improve algorithmic robustness.

Main Results:

• Wasserstein distance-based methods (exact and approximate) outperformed state-of-the-art algorithms on synthetic data with spread class information.
• The HABiC classifier demonstrated consistently higher accuracy in predicting clinical and biological outcomes from transcriptomics data.
• Comparison with Euclidean distance-based classifiers and neural network approximations highlighted the efficacy of the proposed method.

Conclusions:

• The developed Wasserstein distance-based algorithm offers a more precise and robust approach for omics data analysis in precision oncology.
• HABiC shows significant potential for improving clinical outcome prediction accuracy.
• The findings suggest a promising direction for advancing ML applications in cancer research and treatment.