An eigenvector is a direction associated with a linear transformation. In principal component analysis (PCA), eigenvectors of the covariance matrix identify directions along which standardized molecular features vary most; their eigenvalues measure the variance explained. section 6.2.3
PCA uses eigenvectors to create a lower-dimensional representation:
- Standardize features so differing scales contribute equally. section 6.2.3
- Compute the covariance matrix and its eigenvectors and eigenvalues. section 6.2.3
- Sort eigenvectors by descending eigenvalue, retain the first k, and represent each sample by coordinates along them. section 6.2.3
Choosing informative directions
A dimension whose values barely vary contributes little information, while a direction with substantial variation is more useful for compression. PCA retains the leading eigenvectors and chooses k using eigenvalue, explained-variance, or scree-plot criteria.




