AbstractBatch effect correction has been recognized to be indispensable when integrating single-cell RNA sequencing (scRNA-seq) data from multiple batches. State-of-the-art methods ignore single-cell cluster label information, but such information can improve effectiveness of batch effect correction, particularly under realistic scenarios where biological differences are not orthogonal to batch effects. To address this issue, we propose SMNN for batch effect correction of scRNA-seq data via supervised mutual nearest neighbor detection. Our extensive evaluations in simulated and real datasets show that SMNN provides improved merging within the corresponding cell types across batches, leading to reduced differentiation across batches over MNN, Seurat v3, and LIGER. Furthermore, SMNN retains more cell type-specific features, partially manifested by differentially expressed genes identified between cell types after SMNN correction being biologically more relevant, with precision improving by up to 841%.Key PointsBatch effect correction has been recognized to be critical when integrating scRNA-seq data from multiple batches due to systematic differences in time points, generating laboratory and/or handling technician(s), experimental protocol, and/or sequencing platform.Existing batch effect correction methods that leverages information from mutual nearest neighbors across batches (for example, implemented in SC3 or Seurat) ignore cell type information and suffer from potentially mismatching single cells from different cell types across batches, which would lead to undesired correction results, especially under the scenario where variation from batch effects is non-negligible compared with biological effects.To address this critical issue, here we present SMNN, a supervised machine learning method that first takes cluster/cell-type label information from users or inferred from scRNA-seq clustering, and then searches mutual nearest neighbors within each cell type instead of global searching.Our SMNN method shows clear advantages over three state-of-the-art batch effect correction methods and can better mix cells of the same cell type across batches and more effectively recover cell-type specific features, in both simulations and real datasets.