POIBM: batch correction of heterogeneous RNA-seq datasets through latent sample matching

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput technologies like RNA sequencing are crucial for understanding complex diseases, including cancers.
• Technical factors can introduce batch effects into RNA-seq data, masking biologically relevant patterns.
• Existing batch correction methods often require specific experimental designs or phenotypic labels, which are not always available for patient samples.

Purpose of the Study:

• To develop an RNA-seq batch correction method that does not require phenotypic labels or special experimental designs.
• To evaluate the performance of the proposed method against existing approaches.
• To demonstrate the utility of batch-corrected data for cancer research.

Main Methods:

• Developed POIBM (Probabilistic Orthogonal Iterative Batch-effect Mover), a novel RNA-seq batch correction method.
• POIBM learns virtual reference samples directly from the data without using phenotypic information.
• Evaluated POIBM using a breast cancer cell line dataset and The Cancer Genome Atlas (TCGA) RNA-seq data.

Main Results:

• POIBM demonstrated superior or comparable performance to existing methods on a breast cancer cell line dataset, despite being phenotype-blind.
• Analysis of TCGA RNA-seq data revealed widespread batch effects across numerous cancer types.
• POIBM successfully identified true biological replicates in stomach adenocarcinoma data.
• Integration of POIBM-corrected data improved cancer subtyping in endometrial carcinoma.

Conclusions:

• POIBM offers an effective solution for unbiased batch effect correction in RNA-seq data from heterogeneous populations.
• The method facilitates the discovery of genuine biological signals obscured by technical noise.
• Applying POIBM to large cancer datasets can enhance downstream analyses, such as cancer subtyping and biomarker discovery.