Integrating machine learning and pathway analysis to decode biological insights from the TCGA_BRCA dataset

Description

1. In this repository I intend to show how machine learning classification methods and pathway analysis can lead to decoding biological insights into breast cancer.
2. The TCGA_BRCA dataset can be downloaded following the R script uploaded in this repository. This dataset consists of the transcriptome profile ( Gene expression quantification ) of differnet patients with breast cancer, more details about it can be found at this link: https://portal.gdc.cancer.gov/projects/TCGA-BRCA. The patients are classified based on two differnet types of cancer: "Primary tumor" and "Metastastic", and normal patients: "Solid Tissue Normal"
3. Three machine learning classification methods: Random Forest, Support Vector Machine, and Logistic regression, were used to predict the types of tissues among the patients. Then, pathway analysis was performed to study the biological insights of the dataset.

Scripts

The scripts used can be found in this repository. A more detailed explanation of each script can be found inside each one.

1. Downloading Cancer dataset: R script used to query, download, and save the TCGA_BRCA dataset into a CSV file.
2. Machine learning classification_TCGA_BRCA: Jupyter Notebook ( Python ) script used to perform machine learning methods in the dataset. A more detailed explanation can be found in it.
3. Pathway analysis of the most relevant genes: Jupyter Notebook ( Python ) script was used to obtain the gene importance ( features ) used by the classification methods and then performed a pathway analysis.

Results

1. Random Forest and Support Vector machines were shown to have a better performance than Logistic regression in general. More than 90 % of the samples, from the first two approaches, were predicted correctly.
2. Using Random Forest, the most 20 representative genes, in the dataset, were found, and pathway analysis revealed metabolic and molecular pathways that might be altered in each one of the types of cancer. Just two gene sets were used in this study: GO_Biological_process_2023 and GO_Molecular_Functions_2023. Plots are displayed in this script showing the pathways associated with those genes.

Future directions

1. For better classification performance, neural networks will be used and added to this repository.
2. Validation of this model with other breast cancer datasets will be performed and added to this repository.
3. Finally, a multi-omics integration will be performed and will be added to this repository.