Biomarker Discovery and Prediction

Services:

At BioMath Solutions we are experienced in development of predictive biomarkers and classifier algorithm design using the latest machine learning techniques [References].

The key development features are:

• Feature selection

In the context of gene-expression classification, the process of selecting a subset of relevant variables (gene, proteins, est.) from an enormous set of possible variables to build the classification algorithm. This is the first step in the biomarker development process, the selection of the variables that best discriminate between groups, for example, those genes that distingue normal and cancerous tissue.

• Classification

Supervised classification is a machine learning technique to deduce a mathematical function to group similar objects based on training data. The process of taking your biomarker (gene, protein, est.) and development of the function to mathematically distingue between groups. In high-throughput Genomics applications for example, including diagnosis and prognosis, the objective is to define functions that can discriminate different phenotypes based on the expression of a set of genes. The design of such classifiers involves the selection of optimal classification rules, estimation of error rates, and the best use of available samples.

• Error estimation

Many biomarker projects fail because of improper performance estimate. It is critical during the biomarker development process to use the correct error estimation at each step of the process, failure to do so will result in either an over or under optimist performance of ones biomarker.