Unbiased integration of single cell transcriptomes using a linear hybrid method

Single cell transcriptomic approaches are becoming mainstream, with replicated experiments performed in the same single cell platform being more common. Methods that enable to integrate these datasets by removing batch effects while preserving biological information are required. Here we present Canek, a method that, leveraging information from mutual nearest neighbors, combines a local linear correction with a cell-specific non-linear correction using fuzzy logic. Using a combination of real and synthetic datasets we show that Canek corrects batch effects while introducing the least amount of bias when compared with competing methods. Canek is computationally efficient, being able to integrate single cell transcriptomes for thousands of cells from replicated experiments.