Lense: Optimizing data preprocessing in single-cell omics using LLMs.

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Genomics

Background:

• Data preprocessing is essential for single-cell omics studies.
• Existing preprocessing pipelines struggle with diverse datasets, particularly from new technologies like spatial transcriptomics.
• Optimal preprocessing is often dataset-specific and requires manual tuning.

Purpose of the Study:

• To introduce Lense, an automated method for selecting optimal data preprocessing pipelines in single-cell omics.
• To improve the robustness and efficiency of preprocessing for diverse omics datasets.
• To reduce the need for manual intervention in data preprocessing.

Main Methods:

• Lense utilizes a language-model-guided approach to evaluate preprocessing pipeline variants.
• It compares low-dimensional representations visualized through plots to identify optimal preprocessing steps.
• The method is integrated with the Seurat analysis package for seamless workflow implementation.

Main Results:

• Lense automatically selects optimal preprocessing strategies, outperforming default pipelines on diverse datasets.
• The method demonstrates improved preprocessing robustness, especially for emerging platforms like spatial transcriptomics.
• Integration with Seurat streamlines the analysis workflow without manual parameter tuning.

Conclusions:

• Lense offers an automated and robust solution for data preprocessing in single-cell omics.
• The language-model-guided approach enhances the reliability of omics data analysis.
• Lense is a valuable tool for researchers working with diverse and complex omics datasets, including spatial transcriptomics.