On the construction of medical test systems using greedy algorithm and support vector machine

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Medicine

Background:

• Genomic classifiers are crucial for personalized medicine, particularly in oncology.
• Current methods often rely on extensive biological and medical expertise.
• There is a need for data-driven approaches to genomic classifier development.

Purpose of the Study:

• To formalize the problem of selecting parameters and constructing genomic classifiers using machine learning.
• To propose and test a novel method for building genomic classifiers independent of biological knowledge.
• To evaluate the performance of the developed classifier against established breast cancer tests.

Main Methods:

• Formalized problem statement for genomic classifier construction.
• Application of mathematical machine learning methods.
• Utilized microarray datasets with genome-wide transcriptome data from estrogen-positive breast tumors.
• Developed a classifier based on the expression of 12 genes.

Main Results:

• A novel method for constructing genomic classifiers was successfully developed and tested.
• The proposed method achieved classification quality comparable to commercial tests like OncotypeDX and MammaPrint.
• The classifier demonstrated effectiveness using only gene expression data from microarray datasets.

Conclusions:

• Machine learning methods can effectively construct genomic classifiers without prior biological or medical knowledge.
• The developed 12-gene classifier shows promise as a cost-effective and efficient diagnostic tool for estrogen-positive breast cancer.
• This approach offers a new paradigm for developing medical test systems based on genomic data.