12. Statistics for RNA-seq

Which statements regarding unsupervised learning methods for gene expression are correct?

1) A principal Component Analysis (PCA) identifies significantly differently expressed genes -> wrong; these learning methods are not hypothesis testing methods, so they do not spit out a p-value

2) The number of principal components of a PCA is equal to the number of dimensions in the input data matrix-> correct (PCA just resort the data matrix)

3) In a Principal Component Analysis (PCA) the first principal component (PC1) always explains more or equal variation than the second principal component (PC2) -> correct

4) PCA, t-SNE and UMAP identify clusters and assign p-values to cluster differences -> wrong, these learning methods are not hypothesis testing methods, so they do not spit out a p-value

5) An advantage of t-SNE versus PCA is that local cluster structures are better visible -> correct

6) Hierarchical clustering of gene expression can be done with different measures of similarity, such as the Pearson correlation coefficient or the Euclidian distance -> correct

7) Applying k-means clustering to gene expression data results in k clusters of genes and/or samples whereas k is a parameter that needs to bet set -> correct

Which statements regarding pathway/Gene set analyses in the context of gene expression analyses is correct?

1) In a Gene Ontology, a gene is usually annotated in several categories/GO terms -> correct

2) The Gene Ontology is a method to detect enriched pathways in differentially expressed genes -> wrong, GO is not a method, it’s a database that stores gene annotation; the method is (should be) Gene set enrichment analysis, which uses GO for that matter

3) The Gene Ontology consortium annotates genes according to three main domains: Molecular Function, Biological Process and Cellular component -> correct

4) If a pathway is said to be differently expressed between two groups, it usually means that all genes in the pathway are differently expressed between the two groups -> wrong; if a pathway is affected, its just enriched, its more affected than you expect on average -> doesn’t mean all are higher expressed, just that a higher fraction is differently expressed (if one e.g. has a big group of genes than even just a slightly higher fraction of differently expressed genes will be significant; if it’s a small group, I need a big difference)

5) A more significant enrichment of differently expressed genes in GO group 1 than in GO group 2 also means that a higher fraction of genes are differently expressed in GO group 1 than in GO group 2 -> wrong, e.g. if the GO have different sizes -> (if one e.g. has a big group of genes than even just a slightly higher fraction of differently expressed genes will be significant; if it’s a small group, I need a big difference)