Conceptual

Principal Component Analysis

An orthogonal linear projection onto directions of maximal variance, computed from the eigenvectors of the covariance matrix. In transcriptomics it compresses tens of thousands of genes into tens of components before any neighbourhood graph or embedding is built.