Biomedical Machine Learning

Gene Network Estimation

Figure 2-1. Estimated gene network by the two-stage adaptive lasso.

In recent years, we have been facing an explosion of accumulated gene expression data, due to the low cost of genome profiling resulting from the sequencing technology development. Along with such gene expression big data, there have been many studies using the data for complex problems in bio-medicine.

Gene network estimation is one of the tasks, which can be used for the drug discovery problem and estimating biological pathway. In order to handle the hyper-dimensional properties in gene expression, we are studying penalization and regularization methods for machine learning methods.

Survival Estimation

Figure 2-2. Gene network-based survival estimation. (Left) Estimated risk-gene network for low grade glioma (Right) Comparison of estimated risk between the conventional method and the proposed method.

Prediction of patients’ future risk is a crucial problem since we can decide in advance whether to apply costly but powerful treatments, or low-priced and less side-effect treatments. To address such a problem, we proposed a gene network-based risk estimation method.