Projected t-SNE for batch correction.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Visualization

Background:

• Low-dimensional representations like t-SNE are crucial for visualizing high-dimensional biomedical data.
• Batch effects can obscure underlying biological structures in these datasets.
• Accurate interpretation of complex biological data requires effective batch effect removal.

Purpose of the Study:

• To develop a novel procedure for estimating t-distributed Stochastic Neighbor Embedding (t-SNE) embeddings that are not influenced by batch effects.
• To improve the visualization and interpretation of high-dimensional biomedical data by mitigating batch effects.
• To provide an efficient and computationally fast algorithm for batch-corrected t-SNE.

Main Methods:

• The proposed method utilizes linear algebra and constrained optimization techniques.
• The algorithm directly estimates t-SNE embeddings, incorporating batch effect correction.
• The approach is designed for efficiency and fast computation in high-dimensional settings.

Main Results:

• The procedure successfully removes multiple batch effects from t-SNE embeddings in simulated single-cell transcription profiling data.
• Fundamental information on cell types is retained after batch effect correction.
• Application to mouse medulloblastoma single-cell gene expression data demonstrated removal of experimental batches (e.g., mouse ID, date) while preserving key cell populations.

Conclusions:

• The novel algorithm effectively removes batch effects from t-SNE visualizations.
• The method preserves essential biological information, aiding in the accurate identification of cell types and structures.
• This approach significantly enhances the utility of t-SNE for analyzing complex, high-dimensional biomedical datasets.