iEMNN: An Iterative Integration Method for Single-Cell Transcriptomic Data Based on Network Similarity Enhancement and Mutual Nearest Neighbors

With the widespread application of single-cell RNA sequencing technology, integrating different batches of data has become a crucial step. Batch effects arise from non-biological variations such as different sequencing batches, sequencing protocols, sequencing depths, and so on. Batch effects introduce systematic biases and confound biological variations of interest, which have a detrimental impact on the validity of study findings. Eliminating batch effects can increase comparability and repeatability, prevent bias and confusion, and improve data quality and consistency. Currently, a great deal of batch effect removal techniques for single-cell transcriptome data has been developed based on finding mutual nearest neighbors (MNNs) across batches, the accuracy of which influences greatly the effect of data correction and the subsequent integrative analysis. To enhance the identification of MNNs, we propose an iterative integration method, called iEMNN, for single-cell transcriptome data by utilizing network similarity enhancement. iEMNN facilitates the detection of similar cells while separating distinct cells. iEMNN applies multiple iterations to improve the effectiveness of the integration process and can perform data correction in both high-dimensional and low-dimensional spaces. Through systematic experiments and comparisons with the existing methods, we demonstrate that iEMNN outperforms other approaches in improving batch correction removal and enhancing clustering performance across different scenarios. This study provides a new perspective and an effective solution for the field of single-cell transcriptomic data integration, with potential significance for a deeper understanding of cellular heterogeneity and dynamic changes.