|
Abstract:
|
Batch effects, sources in variation across conditions unrelated to biological or scientific variables in a study, can confound true associations in the analysis of high throughput data. A common goal of many high throughput experiments is to quantify the relationships between measured features through network inference. However, if not corrected for, batch effects can disrupt network inference algorithms, increase false edge discoveries, and reduce replicability of inferred networks. Additionally, true relationships may be missed. We discuss proposed methods to correct for confounders such as batch effects before implementing network inference methods. We show that when the batch variable is unknown, principal component correction of feature-level measurements before applying network inference algorithms reduces false discoveries in inferred graphs and increases network preservation.
|
