Publicação
Analysis of human metabolic models using machine learning techniques
| Resumo: | In this work we have analyzed published cancer genome-scale metabolic models from Biomodels database. We have investigated if these metabolic models reveal some information specific to the patients. For this goal we have constructed DataFrames using the reactions, metabolites and clinical information. For the cases of Liver and Breast cancer models we have researched if unsupervised machine learning techniques were able to clas sify data into different phenotype groups. We have used K-means and Hierarchical clustering methods for this reason. We have observed that for Liver cancer the patients cluster into sub-groups with significant different survival times and tumor stages. For Breast cancer there is still some differences but not as significant as the Liver cancer. We have also used supervised machine learning techniques like Support Vector Machine, Random Forest and K-Nearest Neighbor to analyze if we can predict some features such as tumor stage, days to death and vital status. We have used imbalanced techniques such as SMOTE and ADASYN. We have concluded that we can predict tumor stage better than days to death and vital status features. This work demonstrates that metabolic modeling can be a helpful tool for predicting patient-specific features along the goal to the personalized medicine. |
|---|---|
| Autores principais: | Morim, Pedro Miguel Serra |
| Assunto: | Biomodels database Machine learning Liver cancer Breast cancer TCGA data Human metabolic models Cancro do fígado Cancro da mama Modelos metabólicos humanos |
| Ano: | 2020 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | In this work we have analyzed published cancer genome-scale metabolic models from Biomodels database. We have investigated if these metabolic models reveal some information specific to the patients. For this goal we have constructed DataFrames using the reactions, metabolites and clinical information. For the cases of Liver and Breast cancer models we have researched if unsupervised machine learning techniques were able to clas sify data into different phenotype groups. We have used K-means and Hierarchical clustering methods for this reason. We have observed that for Liver cancer the patients cluster into sub-groups with significant different survival times and tumor stages. For Breast cancer there is still some differences but not as significant as the Liver cancer. We have also used supervised machine learning techniques like Support Vector Machine, Random Forest and K-Nearest Neighbor to analyze if we can predict some features such as tumor stage, days to death and vital status. We have used imbalanced techniques such as SMOTE and ADASYN. We have concluded that we can predict tumor stage better than days to death and vital status features. This work demonstrates that metabolic modeling can be a helpful tool for predicting patient-specific features along the goal to the personalized medicine. |
|---|