The problem. Real single-cell studies combine many samples, batches, technologies, even species. Naively pooling them, batch effects dominate: cells cluster by where they were run, not by cell type. But over-correct and you scrub out real biological differences. Integration has to align shared cell states across datasets while preserving genuine variation.

The idea. Seurat v3 introduces “anchors”: pairs of cells across datasets identified (via shared nearest neighbours in a common space) as the same biological state. These anchors define a transformation that harmonises datasets, letting you transfer labels from a reference onto a query or co-embed disparate datasets. It’s a principled, correspondence-based approach to batch correction.

Why it matters. Seurat is the R half of the single-cell world whose Python half is Scanpy — knowing both is knowing the field’s two default toolkits. The anchor idea also connects to scANVI’s label transfer (day 4): reference-to-query annotation is a recurring theme, solved here geometrically. Integration is the step that makes atlas-scale single-cell analysis possible at all.

Verdict. Foundational and enormously influential; anchors became a standard integration paradigm. Its corrections need care — aggressive integration can hide real differences — but the framework is central. Read it as Scanpy’s sibling and the batch-effect workhorse.