Dimensionality reduction and clustering
PCA finds directions of large variation; a two-dimensional PCA plot is a projection that discards some information.
Loading video…
PCA finds directions of large variation; a two-dimensional PCA plot is a projection that discards some information. Clustering groups samples by a chosen distance and algorithm. Apparent separation can come from batch, scaling or random variation. Use labels for interpretation after fitting an exploratory method, then test biological hypotheses separately.
Worked example
If PC1 separates sequencing batches rather than treatments, the strongest signal is technical. Color the same PCA plot by batch and treatment before claiming disease groups differ.