NormAE: Deep Adversarial Learning Model to Remove Batch Effects in Liquid Chromatography Mass Spectrometry-Based Metabolomics Data

Abstract

Untargeted metabolomics based on liquid chromatography−mass spectrometry is affected by nonlinear batch effects, which cover up biological effects, result in nonreproducibility, and are difficult to be calibrate. In this study, we propose a novel deep learning model, called Normalization Autoencoder (NormAE), which is based on nonlinear autoencoders (AEs) and adversarial learning. An additional classifier and ranker are trained to provide adversarial regularization during the training of the AE model, latent representations are extracted by the encoder, and then the decoder reconstructs the data without batch effects. The NormAE method was tested on two real metabolomics data sets. After calibration by NormAE, the quality control samples (QCs) for both data sets gathered most closely in a PCA score plot (average distances decreased from 56.550 and 52.476 to 7.383 and 14.075, respectively) and obtained the highest average correlation coefficients (from 0.873 and 0.907 to 0.997 for both). Additionally, NormAE significantly improved biomarker discovery (median number of differential peaks increased from 322 and 466 to 1140 and 1622, respectively). NormAE was compared with four commonly used batch effect removal methods. The results demonstrated that using NormAE produces the best calibration results.

Addressing the batch effect issue for LC/MS metabolomics data in data preprocessing

“…Given there were six batches, the between-batch detection proportions meant we required a feature to be initially detected in at least 1, 2, 3, 4, 5, or 6 batches, respectively. For the traditional apLCMS procedure, we set the detection threshold (number of samples) at 30 XCMS IPO_4: CentWave parameters: same as XCMS IPO_3; peak grouping parameters: bw = 22, mzwid = 0.018; Loess parameters: missing = 1, extra = 3, span = 0.2, smooth = "loess", family = "gaussian".…”

reComBat: batch-effect removal in large-scale multi-source gene-expression data integration

“…Recently, nonlinear models, often based on neural autoencoders, have gained popularity (e.g. normAE [40], AD-AE [10], or scGEN [32]). Most models aim to find a batch-effect-free latent space representation of the data via adversarial training.…”

“…Additionally, desired biological variation (referred to in this paper as " (experimental) design") between different independent experiments needs be conserved in any algorithm which aims to remove the batch effects. Although a range of batch correction algorithms has previously been suggested [46,28,40,8], only a small subset of these remains applicable in this large-scale setting. In particular, most previous algorithms cannot incorporate high-dimensional experimental design information.…”

Deep learning based multi-batch calibration for classification in various omics

“…With the evolution of deep network architecture, there are more and more breakthroughs in GANs 14 in the past three years. A typical method applied to this field is the NormAE 15 developed by Rong et al in 2020. Its basic idea lies in constructing an adversarial training procedure between a nonlinear AE to remove batch effects and a discriminator to distinguish the source of domain based on the latent space.…”

“…b Comparison of classification accuracy with multiple source batches for training and only one target batch for testing. Note that "Recon_T" denotes an ablation experiment that reconstruct all target batches.Afterwards, we select several latest and most representative tools including ComBat 17 , NormAE15 , BERMUDA4 , and DESC 5 for further comparison. Accuracy of cross-batch prediction in the sample level is utilized to assess the effectiveness of each method.…”