Toward universal immunofluorescence normalization for multiplex tissue imaging with UniFORM.

Area of Science:

• Computational Biology
• Bioinformatics
• Image Analysis

Background:

• Multiplex tissue imaging (MTI) generates complex datasets requiring robust normalization.
• Existing MTI normalization methods struggle to balance batch effect removal with biological signal preservation.

Purpose of the Study:

• To introduce UniFORM, a novel non-parametric Python pipeline for MTI data normalization.
• To address technical variations at both feature and pixel levels in MTI data.
• To enhance the accuracy and biological interpretability of MTI analyses.

Main Methods:

• UniFORM utilizes an automated rigid landmark registration method tailored for MTI data characteristics.
• The pipeline operates without prior distributional assumptions, accommodating unimodal and bimodal patterns.
• It aligns negative populations to remove technical variation while preserving positive population expression patterns.

Main Results:

• UniFORM demonstrated superior performance across three MTI platforms compared to existing methods.
• Improved marker distribution alignment and positive population preservation were observed.
• Enhanced k-nearest neighbor batch effect test (kBET) and silhouette scores indicate better data quality.

Conclusions:

• UniFORM effectively mitigates batch effects in MTI data while maintaining biological signal fidelity.
• The pipeline supports more coherent downstream analyses, including UMAP and Leiden clustering.
• UniFORM's scalable design and optional fine-tuning mode offer broad applicability for MTI data normalization.