What are common causes of batch effects in RNA-seq?

Why

Start with the sample sheet. Record specimen source, donor, extraction method, RNA integrity, preparation date, reagent lot, read depth, alignment rate and sequencing lane. Then inspect PCA or related exploratory plots colored separately by condition and batch.

Suppose all controls were prepared on Monday and all treated samples on Friday. Even if downstream software detects a difference, it cannot reliably allocate that difference to treatment versus day, because the design lacks overlap. Prevent this by randomizing sample order and ensuring each processing batch includes both conditions.

Where independent batches and condition levels coexist, model batch as a covariate or use a paired design when appropriate. Avoid blind correction before diagnostic plots, because an aggressive algorithm can remove biological signal. Biological variables such as cell mixture or donor ancestry may correlate with batch and need a specific scientific decision, not automatic deletion.