TCGA data mining for establishing the relationship between sub-branches of Wnt signalling in colorectal cancer (CRC)

Project completed during my BSc in Biochemistry at the University of Southampton, 2018-2019.

Data extraction

• produce a library of Wnt signalling components and their associated genes of proteins
• search for CRC gene expression data (e.g. RNAseq) in publicly-available datasets (e.g. TCGA)

Data pre-processing

• ensure a fair data comparison (data was normalised using RSEM - RPKM modelled to account for isoform abundances)
• address the problem of missing data

Data analysis

• uses gplots, RColorBrewer and dunn.test R packages

1. EXPLORATORY DATA ANALYSIS
   • check distribution of data
2. UNSUPERVISED MACHINE LEARNING
   • hierarchical clustering (no prior information was known about the groups)
   • Pearson correlation heatmap (for computing distances using complete linkage)
3. CLUSTER ANALYSIS FOR SIGNIFICANT GROUPS
   • Kruskal-Wallis test for comparing groups
   • post-hoc Dunn's test with Benjamini-Hochberg p-value correction for finding which group was different
   • hypergeometric test for the probabilities of randomly picking the samples from a certain category
4. FURTHER ANALYSIS
   • gene set enrichment (PANTHER)
   • PPIs network (STRING)