Authors:
Alaa Bakry, Andreas Eberhard, Debasmita Asari, Chou Yi Hsu, Shahid Ullah, Lilianne Dominguez Céspedes

Addresses:
Center for Bioinformatics, Saarland University, Saarbrücken, Saarland, Germany. Department of Pharmacy, Chia Nan University of Pharmacy and Science, Tainan City, Taiwan. Department of Bioinformatics and Biochemistry, S. Khan Lab, Mardan, Khyber Pakhtunkhwa, Pakistan. Department of Clinical Bioanalysis, Filial of Medical Sciences of Holguín, Avenida Lenin, Esquina Aguilera, Holguín, Cuba.

Abstract:

Batch effects (BEs) systematic, non-biological discrepancies introduced when samples are processed across different locations, platforms, or times remain a persistent obstacle to the integrative analysis of RNA-Seq datasets. Reliable correction methods must remove technical variation without eroding genuine biological signal, yet a standardised way to benchmark these methods against data with known, controllable BE magnitude is lacking. Researchers introduced synthetic batch effects into two public RNA-Seq datasets using three complementary strategies additive random noise scaled to a fraction of the standard deviation, row/column mean augmentation, and cluster-mean augmentation following k-means partitioning and then applied four established BE-removal methods (ComBat, Surrogate Variable Analysis (SVA), BEclear, and Mutual Nearest Neighbors (MNN)) to recover the original signal. All four methods reduced batch-associated variance relative to the uncorrected data. MNN achieved the strongest correction, yielding the lowest post-correction correlation between principal components (−1.316×10-5). At the same time, ComBat and SVA showed comparable, moderate performance, and BEclear performed well but at higher computational cost. Researchers found no existing software tool purpose-built for injecting synthetic batch effects, motivating our three custom introduction strategies.

Keywords: Non-Biological Discrepancies; Persistent Obstacle; Integrative Analysis; RNA-Seq Datasets; Eroding Genuine Biological Signal; Surrogate Variable Analysis; Mutual Nearest Neighbours.

Received on: 27/09/2024, Revised on: 22/12/2024, Accepted on: 27/04/2025, Published on: 15/09/2026

DOI: 10.69888/FTSASS.2026.000768

FMDB Transactions on Sustainable Applied Sciences, 2026 Vol. 3 No. 2, Pages: 62-73

• Views : 43
• Downloads : 8

Download PDF