Data Normalization Strategies Research Articles

Abstract LB012: Evaluating computational approaches for CPTAC pan-cancer cross-cohort protein expression comparison

Abstract Recently, the National Cancer Institute’s Clinical Proteomic Tumor Analysis Consortium (CPTAC) has generated harmonized genomic, transcriptomic, proteomic, and clinical data for >1000 tumors in 10 cohorts to facilitate pan-cancer discovery research. However, protein expression comparison across CPTAC cohorts remains challenging due to non-uniform missing data and protein expression distribution patterns across tumor types. Here we present our efforts to evaluate various missing data handling and normalization strategies to generate a normalized pan-cancer protein expression dataset. First, we developed a novel algorithm to select robustly expressed proteins in tumors in any of the CPTAC cohorts, Second, we applied a cohort hybrid imputation approach to protein abundance values from FragPipe within each cohort based on protein expression distribution patterns. Third, we calculated iBAQ using protein abundance value and applied global quantile normalization or smooth quantile normalization methods. To assess if our missing data imputation and normalization strategy affected downstream analyses, we compared the fold change in differential protein expression between tumor and matched normal for each cohort using non-normalized, global quantile normalized and smooth quantile normalized protein iBAQ values. Our results demonstrate a strong correlation in fold change between global quantile normalized data and non-normalized data (Pearson r = 0.97 (ccRCC), r = 0.96 (COAD), r = 0.99 (LUAD) and r = 0.99 (LSCC)). Similar results were observed comparing smooth quantile normalized data to non-normalized data (Pearson r = 1.00 for ccRCC, COAD, LUAD, and LSCC), indicating both normalization methods retained biological differences between tumor and matched normal tissues within cohorts. Lastly, we identified several proteins (ERAP2, CA9, GSTM3, MX1, STAT1) whose protein and RNA expression were highly correlated across eight CPTAC cohorts (r > 0.7 for COAD, BRCA, LUAD, ccRCC, PDAC, UCEC, HNSCC, and LSCC). We then compared their protein expression rank across CPTAC cohorts with their RNA expression rank across corresponding TCGA cohorts. Specifically, median log2(iBAQ) of CPTAC and median log2(TPM) of TCGA are calculated for those proteins within indications, then indications are ranked by median log2(iBAQ) and median log2(TPM) in CPTAC and TCGA, respectively. Weighted rank correlation was used to measure rank agreement. Global quantile normalization has the highest rank correlation (weighted rank correlation between 0.597 to 0.931) compared to smooth quantile normalization or without normalization. These results suggest that combination of cohort hybrid imputation and global quantile normalization is a reasonable approach to generate a normalized CPTAC pan-cancer protein dataset that could be leveraged to interrogate protein expression across different cancer types. Citation Format: Jixin Wang, Xiaowen Tian, Wen Yu, John Bullen, Elaine Hurt, Wenyan Zhong. Evaluating computational approaches for CPTAC pan-cancer cross-cohort protein expression comparison [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB012.

Evaluation of normalization strategies for GC-based metabolomics.

For many samples studied by GC-based metabolomics applications, extensive sample preparation involving extraction followed by a two-step derivatization procedure of methoximation and trimethylsilylation (TMS) is typically required to expand the metabolome coverage. Performing normalization is critical to correct for variations present in samples and any biases added during the sample preparation steps and analytical runs. Addressing the totality of variations with an adequate normalization method increases the reliability of the downstream data analysis and interpretation of the results. Normalizing to sample mass is one of the most commonly employed strategies, while the total peak area (TPA) as a normalization factor is also frequently used as a post-acquisition technique. Here, we present a new normalization approach, total derivatized peak area (TDPA), where data are normalized to the intensity of all derivatized compounds. TDPA relies on the benefits of silylation as a universal derivatization method for GC-based metabolomics studies. Two sample classes consisting of systematically incremented sample mass were simulated, with the only difference between the groups being the added amino acid concentrations. The samples were TMS derivatized and analyzed using comprehensive two-dimensional gas chromatography coupled to time-of-flight mass spectrometry (GC × GC-TOFMS). The performance of five normalization strategies (no normalization, normalized to sample mass, TPA, total useful peak area (TUPA), and TDPA) were evaluated on the acquired data. Of the five normalization techniques compared, TUPA and TDPA were the most effective. On PCA score space, they offered a clear separation between the two classes. TUPA and TDPA carry different strengths: TUPA requires peak alignment across all samples, which depends upon the completion of the study, while TDPA is free from the requirement of alignment. The findings of the study would enhance the convenient and effective use of data normalization strategies and contribute to overcoming the data normalization challenges that currently exist in the metabolomics community.

Normalization approaches for extracellular vesicle-derived lipidomic fingerprints – A human milk case study

Human milk (HM) extracellular vesicles (EVs) are nano-sized, cell-derived particles sheathed in a lipid bilayer that encase specific cargo for delivery from mother to infant. The aim of this study was to expand our understanding of the lipidomic fingerprint of HM-EVs, with a specific focus on the impact of data normalization using simulated and experimental data obtained from the analysis of HM samples from mothers of preterm (N = 5) and term infants (N = 5), and a pool of donor human milk from 20 mothers (before and after pasteurization). EVs were isolated by multi-stage ultracentrifugation and characterized in terms of total protein content, total particle count and size, surface tetraspanin profile and protein markers, and morphology. Lipidomic analysis after single-phase extraction was performed by liquid chromatography mass spectrometry (LC-MS). The effect of widely used data normalization strategies (i.e., sample volume, particle count, protein content, total lipids signal) was compared. Results show that for the selection of the optimum normalization approach, the specific study aims, as well as the purity and homogeneity of size distribution of EV isolates should be considered. While normalization attending the particle number can be useful for between sample comparisons in EV populations with similar particle size, normalization to total lipid content is preferred when lipid contamination is encountered. Our findings exemplify the need for guidance with respect to data processing in LC-MS-based lipidomics studies of EVs.

Variability in donor leukocyte counts confound the use of common RNA sequencing data normalization strategies in transcriptomic biomarker studies performed with whole blood

Gene expression data generated from whole blood via next generation sequencing is frequently used in studies aimed at identifying mRNA-based biomarker panels with utility for diagnosis or monitoring of human disease. These investigations often employ data normalization techniques more typically used for analysis of data originating from solid tissues, which largely operate under the general assumption that specimens have similar transcriptome composition. However, this assumption may be violated when working with data generated from whole blood, which is more cellularly dynamic, leading to potential confounds. In this study, we used next generation sequencing in combination with flow cytometry to assess the influence of donor leukocyte counts on the transcriptional composition of whole blood specimens sampled from a cohort of 138 human subjects, and then subsequently examined the effect of four frequently used data normalization approaches on our ability to detect inter-specimen biological variance, using the flow cytometry data to benchmark each specimens true cellular and molecular identity. Whole blood samples originating from donors with differing leukocyte counts exhibited dramatic differences in both genome-wide distributions of transcript abundance and gene-level expression patterns. Consequently, three of the normalization strategies we tested, including median ratio (MRN), trimmed mean of m-values (TMM), and quantile normalization, noticeably masked the true biological structure of the data and impaired our ability to detect true interspecimen differences in mRNA levels. The only strategy that improved our ability to detect true biological variance was simple scaling of read counts by sequencing depth, which unlike the aforementioned approaches, makes no assumptions regarding transcriptome composition.

A study of Bayesian deep network uncertainty and its application to synthetic CT generation for MR-only radiotherapy treatment planning.

The use of synthetic computed tomography (CT) for radiotherapy treatment planning has received considerable attention because of the absence of ionizing radiation and close spatial correspondence to source magnetic resonance (MR) images, which have excellent tissue contrast. However,in an MR-only environment, little effort has been made to examine the quality of synthetic CT images without using the original CT images. To estimate synthetic CT quality without referring to original CT images, this study established the relationship between synthetic CT uncertainty and Bayesian uncertainty, and proposed a new Bayesian deep network for generating synthetic CT images and estimating synthetic CT uncertainty for MR-only radiotherapy treatment planning. A novel deep Bayesian network was formulated using probabilistic network weights. Two mathematical expressions were proposed to quantify the Bayesian uncertainty of the network and synthetic CT uncertainty, which was closely related to the mean absolute error (MAE) in Hounsfield Unit (HU) of synthetic CT. These uncertainties were examined to demonstrate the accuracy of representing the synthetic CT uncertainty using a Bayesian counterpart. We developed a hybrid Bayesian architecture and a new data normalization scheme, enabling the Bayesian network to generate both accurate synthetic CT and reliable uncertainty information when probabilistic weights were applied. The proposed method was evaluated in 59 patients (13/12/32/2 for training/validation/testing/uncertainty visualization) diagnosed with prostate cancer, who underwent same-day pelvic CT- and MR-acquisitions. To assess the relationship between Bayesian and synthetic CT uncertainties, linear and non-linear correlation coefficients were calculated on per-voxel, per-tissue, and per-patient bases. For accessing the accuracy of the CT number and dosimetric accuracy, the proposed method was compared with a commercially available atlas-based method (MRCAT) and a U-Net conditional-generative adversarial network (UcGAN). The proposed model exhibited 44.33 MAE, outperforming UcGAN 52.51 and MRCAT 54.87. The gamma rate (2%/2mm dose difference/distance to agreement) of the proposed model was 98.68%, comparable to that of UcGAN (98.60%) and MRCAT (98.56%). The per-patient and per-tissue linear correlation coefficients between the Bayesian and synthetic CT uncertainties ranged from 0.53 to 0.83, implying a moderate to strong linear correlation. Per-voxel correlation coefficients varied from -0.13 to 0.67 depending on the regions-of-interest evaluated, indicating tissue-dependent correlation. The R2 value for estimating MAE solely using Bayesian uncertainty was 0.98, suggesting that the uncertainty of the proposed model was an ideal candidate for predicting synthetic CT error, without referring to the original CT. This study established a relationship between the Bayesian model uncertainty and synthetic CT uncertainty. A novel Bayesian deep network was proposed to generate a synthetic CT and estimate its uncertainty. Various metrics were used to thoroughly examine the relationship between the uncertainties of the proposed Bayesian model and the generated synthetic CT. Compared with existing approaches, the proposed model showed comparable CT number and dosimetric accuracies. The experiments showed that the proposed Bayesian model was capable of producing accurate synthetic CT, and was an effective indicator of the uncertainty and error associated with synthetic CT in MR-only workflows.

Exploration and optimization of extraction, analysis and data normalization strategies for mass spectrometry-based DNA adductome mapping and modeling

Although interest in characterizing DNA damage by means of DNA adductomics has substantially grown, the field of DNA adductomics is still in its infancy, with room for optimization of methods for sample analysis, data processing and DNA adduct identification. In this context, the first objective of this study was to evaluate the use of hydrophilic interaction (HILIC) vs. reversed phase liquid chromatography (RPLC) coupled to high resolution mass spectrometry (HRMS) and thermal acidic vs. enzymatic hydrolysis of DNA followed by DNA adduct purification and enrichment using solid-phase extraction (SPE) or fraction collection for DNA adductome mapping. The second objective was to assess the use of total ion count (TIC) and median intensity (MedI) normalization compared to QC (quality control), iQC (internal QC) and quality control-based robust locally estimated scatterplot smoothing (LOESS) signal correction (QC-RLSC) normalization for processing of the acquired data. The results demonstrate that HILIC compared to RPLC allowed better modeling of the tentative DNA adductome, particularly in combination with thermal acidic hydrolysis and SPE (more valid models, with an average Q2(Y) and R2(Y) of 0.930 and 0.998, respectively). Regarding the need for data normalization and the management of (limited) system instability and signal drift, QC normalization outperformed TIC, MedI, iQC and LOESS normalization. As such, QC normalization can be put forward as the default data normalization strategy. In case of momentous signal drift and/or batch effects however, comparison to other normalization strategies (like e.g. LOESS) is recommended. In future work, further optimization of DNA adductomics may be achieved by merging of HILIC and RPLC datasets and/or application of 2D-LC, as well as the inclusion of Schiff base stabilization and/or fraction collection in the thermal acidic hydrolysis-SPE sample preparation workflow.

The impact of urine collection method on canine urinary microbiota detection: a cross-sectional study

BackgroundThe urinary tract harbors unique microbial communities that play important roles in urogenital health and disease. Dogs naturally suffer from several of the same urological disorders as humans (e.g., urinary tract infections, neoplasia, urolithiasis) and represent a valuable translational model for studying the role of urinary microbiota in various disease states. Urine collection technique represents a critical component of urinary microbiota research study design. However, the impact of collection method on the characterization of the canine urinary microbiota remains unknown. Therefore, the objective of this study was to determine whether urine collection technique alters the microbial populations detected in canine urine samples. Urine was collected from asymptomatic dogs by both cystocentesis and midstream voiding. Microbial DNA was isolated from each sample and submitted for amplicon sequencing of the V4 region of the bacterial 16 S rRNA gene, followed by analyses to compare microbial diversity and composition between urine collection techniques.ResultsSamples collected via midstream voiding exhibited significantly higher sequence read counts (P = .036) and observed richness (P = .0024) than cystocentesis urine. Bray Curtis and Unweighted UniFrac measures of beta diversity showed distinct differences in microbial composition by collection method (P = .0050, R2 = 0.06 and P = .010, R2 = 0.07, respectively). Seven taxa were identified as differentially abundant between groups. Pasteurellaceae, Haemophilus, Friedmanniella, two variants of Streptococcus, and Fusobacterium were over-represented in voided urine, while a greater abundance of Burkholderia-Caballeronia-Paraburkholderia characterized cystocentesis samples. Analyses were performed at five thresholds for minimum sequence depth and using three data normalization strategies to validate results; patterns of alpha and beta diversity remained consistent regardless of minimum read count requirements or normalization method.ConclusionMicrobial composition differs in canine urine samples collected via cystocentesis as compared to those collected via midstream voiding. Future researchers should select a single urine collection method based on the biological question of interest when designing canine urinary microbiota studies. Additionally, the authors suggest caution when interpreting results across studies that did not utilize identical urine collection methods.

Expanding the Data Normalization Strategy to the MACONT Method for Multi-Criteria Decision Making

The Mixed Aggregation by Comprehensive Normalization Technique (MACONT) is a well-known Multi-Criteria Decision-Making (MCDM) method with significant benefits compared to traditional approaches. The key difference that distinguishes this method from most others is the use of data normalization techniques and aggregation approaches. MACONT uses three different data normalization techniques simultaneously along with two aggregation approaches throughout its evaluation process. This reduces the derivation of evaluation values and enhances the reliability of the final decision results, making the process more precise and convergent. However, the original MACONT emphasizes the integration of multiple normalization techniques of the same type of criteria that might perform badly in some circumstances. This paper proposes combination strategies of six normalization techniques to be coupled with the MACONT to help the normalized data synthetically reflect the original information and solve different types of data, criteria, and alternatives. The proposed approach was applied in four case studies. In all studies, the ranking results were compared with the other MCDM methods, producing the same best alternatives and overcoming the cases when the original MACONT did not work properly.

On the effects of data normalization for domain adaptation on EEG data

In Machine Learning (ML), a well-known problem is the Dataset Shift problem where the data in the training and test sets can follow different probability distributions, leading ML systems toward poor generalization performances. This problem is intensely felt in Brain-Computer Interfaces (BCIs), where bio-signals as Electroencephalographic (EEG) are often used. Indeed, EEG signals are highly non-stationary both over time and between different subjects. To overcome this problem, several solutions are based on transfer learning approaches such as Domain Adaption (DA). In several cases, however, the actual causes of the improvements remain ambiguous. This paper focuses on the impact of data normalization strategies applied together with DA methods. In particular, using SEED, DEAP, and BCI Competition IV 2a EEG datasets, we experimentally evaluated the impact of different normalization strategies applied with and without several well-known DA methods. It results that the choice of the normalization strategy plays a key role on the classifier performances in DA scenarios, and, often, the use of only an appropriate normalization schema outperforms the DA technique. For SEED and BCI Competition IV 2a, a proper normalization strategy alone in a cross-subject context allows to reach accuracy of 81.52±7.26% and 68.52±11.35%, respectively. In a cross-session context, the accuracy of 86.56±8.15% and 67.82±12.48% for SEED and BCI Competition can be reached, respectively. For DEAP, the best cross-subject performance achieved using only normalization was 39.33±14.08%. All these results are comparable with the performance obtained by several well-known DA strategies.

PRSOT: Precipitation Retrieval from Satellite Observations Based on Transformer

Precipitation with high spatial and temporal resolution can improve the defense capability of meteorological disasters and provide indispensable instruction and early warning for social public services, such as agriculture, forestry, and transportation. Therefore, a deep learning-based algorithm entitled precipitation retrieval from satellite observations based on Transformer (PRSOT) is proposed to fill the observation gap of ground rain gauges and weather radars in deserts, oceans, and other regions. In this algorithm, the multispectral infrared brightness temperatures from Himawari-8, the new-generation geostationary satellite, have been used as predictor variables and the Global Precipitation Measurement (GPM) precipitation product has been employed to train the retrieval model. We utilized two data normalization schemes, area-based and pixel-based normalization, and conducted comparative experiments. Comparing the estimated results with the GPM product on the test set, PRSOT_Pixel_based model achieved a Probability Of Detection (POD) of 0.74, a False Alarm Ratio (FAR) of 0.44 and a Critical Success Index (CSI) of 0.47 for two-class metrics, and an Accuracy (ACC) of 0.75 for multi-class metrics. Pixel-based normalization is more suitable for meteorological data, highlighting the precipitation characteristics and obtaining better comprehensive retrieval performance in visualization and evaluation metrics. In conclusion, the proposed PRSOT model has made a remarkable and essential contribution to precipitation retrieval and outperforms the benchmark machine learning model Random Forests.

Implementing a novel deep learning technique for rainfall forecasting via climatic variables: An approach via hierarchical clustering analysis

Variations in rainfall negatively affect crop productivity and impose severe climatic conditions in developing regions. Studies that focus on climatic variations such as variability in rainfall and temperature are vital, particularly in predominant rainfed areas. Forecasting rainfall is very essential in the agriculture sector due to the dependence of many people, while it is very complex to accurately predict rainfall due to its dynamic nature. This study aims to present a deep forecasting model based on optimized (Gated Recurrent Unit) GRU neural network to predict rainfall in Pakistan based on the 30 years of climate data from 1991 to 2020. The climatic variables were first extracted and then fine-tuned by eliminating outliers and extreme values from the data set for precise forecasting. Data normalization strategies were further utilized to adjust numeric values into a standard scale without distorting divergences or losing useful information. The proposed model achieved high prediction accuracy by maintaining minimal Normalized Mean Absolute Error (NMAE) and Normalized Root Mean Squared Error (NRMSE) compared to state-of-the-art rainfall forecasting models. Climatic variables used in the forecasting were evaluated in terms of correlation and regression analysis. The correlation results showed that temperature has a negative association and air quality variables have a positive association with rainfall in each quarter of the year. The second and third quarters of the year showed a high association with rainfall, whereas the air quality variables showed a lesser or no association with rainfall during the first and second quarters of the year. The results further showed a strong association of climatic variables with rainfall for all months of the year. The minimal loss achieved by the proposed model also demonstrated the feasibility of selected variables in precise forecasting of rainfall regardless of volatile climatic conditions.

Comparison of Data Normalization Strategies for Array-Based MicroRNA Profiling Experiments and Identification and Validation of Circulating MicroRNAs as Endogenous Controls in Hypertension.

Introduction: MicroRNAs are small noncoding RNAs with potential regulatory roles in hypertension and drug response. The presence of many of these RNAs in biofluids has spurred investigation into their role as possible biomarkers for use in precision approaches to healthcare. One of the major challenges in clinical translation of circulating miRNA biomarkers is the limited replication across studies due to lack of standards for data normalization techniques for array-based approaches and a lack of consensus on an endogenous control normalizer for qPCR-based candidate miRNA profiling studies. Methods: We conducted genome-wide profiling of 754 miRNAs in baseline plasma of 36 European American individuals with uncomplicated hypertension selected from the PEAR clinical trial, who had been untreated for hypertension for at least one month prior to sample collection. After appropriate quality control with amplification score and missingness filters, we tested different normalization strategies such as normalization with global mean of imputed and unimputed data, mean of restricted set of miRNAs, quantile normalization, and endogenous control miRNA normalization to identify the method that best reduces the technical/experimental variability in the data. We identified best endogenous control candidates with expression pattern closest to the mean miRNA expression in the sample, as well as by assessing their stability using a combination of NormFinder, geNorm, Best Keeper and Delta Ct algorithms under the Reffinder software. The suitability of the four best endogenous controls was validated in 50 hypertensive African Americans from the same trial with reverse-transcription–qPCR and by evaluating their stability ranking in that cohort. Results: Among the compared normalization strategies, quantile normalization and global mean normalization performed better than others in terms of reducing the standard deviation of miRNAs across samples in the array-based data. Among the four strongest candidate miRNAs from our selection process (miR-223-3p, 19b, 106a, and 126-5p), miR-223-3p and miR-126-5p were consistently expressed with the best stability ranking in the validation cohort. Furthermore, the combination of miR-223-3p and 126-5p showed better stability ranking when compared to single miRNAs. Conclusion: We identified quantile normalization followed by global mean normalization to be the best methods in reducing the variance in the data. We identified the combination of miR-223-3p and 126-5p as potential endogenous control in studies of hypertension.

Factors influencing circulating microRNAs as biomarkers for liver diseases.

MicroRNAs (microRNAs) have been implicated to play crucial roles in various liver diseases. Hepatic microRNAs are released in to the circulation in a systematic fashion, and are, therefore, being extensively explored for their role as prognostic or diagnostic markers or therapeutic targets for numerous hepatic diseases. Advantages such as disease- or tissue-specific expression, and ease of detection have implicated circulating microRNAs (c-microRNAs) as the most desirable candidate for biomarker studies. Although several studies have explored c-microRNAs in serum or plasma, consistency of detection of microRNAs remains demanding because of many biological and methodological challenges. Lack of methodological consensus has limited the universal use of c-microRNAs as potential prognostic or diagnostic disease-specific biomarkers. In this review, we have discussed pre-analytical and analytical factors that might affect c-miRNA expression and selection of appropriate data normalization strategy by incorporating endogenous reference microRNAs. This review will aid designing of better approaches and protocols for future diagnostic research on hepatic c-microRNAs.

Normalization of metabolic data to total thymine content and its application to determination of 2-hydroxyglutarate

Our first objective was to develop an approach useful for reliable normalization of 2-hydroxyglutarate (2-HG) intracellular levels. The second objective was to use our data normalization strategy to verify previously published report on the higher d-2-HG level in tumors of colorectal cancer (CRC) patients than in normal colon fragments. We examined various methods of 2-HG level normalization in cell/tissue extracts (number of cells, mass of tissue, total protein). In order to solve the problems with reliable normalization of the 2-HG levels in colon fragments, we proposed a strategy based on relating the concentrations of 2-HG isomers to total thymine concentrations measured by ultra-performance liquid chromatography (UPLC) with UV detection in acid hydrolysates of the cell/tissue extracts. We used a common method of derivatization with diacetyl-l-tartaric anhydride (DATAN) to separate l- and d-2-HG enantiomers. DATAN-derivatized 2-HG was quantitated by UPLC with tandem mass spectrometry (MS/MS) in the selected reaction monitoring (SRM) mode. We observed a linear dependence of the total amount of thymine released from lymphocytes, HCT 116, K562, and PC-3 by acid hydrolysis on their number of cells. Our results showed a significantly higher level of l- and d-2-HG in cancer-free colon than in tumor.

Improving cluster recovery with feature rescaling factors

The data preprocessing stage is crucial in clustering. Features may describe entities using different scales. To rectify this, one usually applies feature normalisation aiming at rescaling features so that none of them overpowers the others in the objective function of the selected clustering algorithm. In this paper, we argue that the rescaling procedure should not treat all features identically. Instead, it should favour the features that are more meaningful for clustering. With this in mind, we introduce a feature rescaling method that takes into account the within-cluster degree of relevance of each feature. Our comprehensive simulation study, carried out on real and synthetic data, with and without noise features, clearly demonstrates that clustering methods that use the proposed data normalization strategy clearly outperform those that use traditional data normalization.

Shrinkage improves estimation of microbial associations under different normalization methods.

Estimation of statistical associations in microbial genomic survey count data is fundamental to microbiome research. Experimental limitations, including count compositionality, low sample sizes and technical variability, obstruct standard application of association measures and require data normalization prior to statistical estimation. Here, we investigate the interplay between data normalization, microbial association estimation and available sample size by leveraging the large-scale American Gut Project (AGP) survey data. We analyze the statistical properties of two prominent linear association estimators, correlation and proportionality, under different sample scenarios and data normalization schemes, including RNA-seq analysis workflows and log-ratio transformations. We show that shrinkage estimation, a standard statistical regularization technique, can universally improve the quality of taxon–taxon association estimates for microbiome data. We find that large-scale association patterns in the AGP data can be grouped into five normalization-dependent classes. Using microbial association network construction and clustering as downstream data analysis examples, we show that variance-stabilizing and log-ratio approaches enable the most taxonomically and structurally coherent estimates. Taken together, the findings from our reproducible analysis workflow have important implications for microbiome studies in multiple stages of analysis, particularly when only small sample sizes are available.

Towards Standardization of Data Normalization Strategies to Improve Urinary Metabolomics Studies by GC×GC-TOFMS.

Urine is a popular biofluid for metabolomics studies due to its simple, non-invasive collection and its availability in large quantities, permitting frequent sampling, replicate analyses, and sample banking. The biggest disadvantage with using urine is that it exhibits significant variability in concentration and composition within an individual over relatively short periods of time (arising from various external factors and internal processes regulating the body’s water and solute content). In treating the data from urinary metabolomics studies, one must account for the natural variability of urine concentrations to avoid erroneous data interpretation. Amongst various proposed approaches to account for broadly varying urine sample concentrations, normalization to creatinine has been widely accepted and is most commonly used. MS total useful signal (MSTUS) is another normalization method that has been recently reported for mass spectrometry (MS)-based metabolomics studies. Herein, we explored total useful peak area (TUPA), a modification of MSTUS that is applicable to GC×GC-TOFMS (and data from other separations platforms), for sample normalization in urinary metabolomics studies. Performance of TUPA was compared to the two most common normalization approaches, creatinine adjustment and Total Peak Area (TPA) normalization. Each normalized dataset was evaluated using Principal Component Analysis (PCA). The results showed that TUPA outperformed alternative normalization methods to overcome urine concentration variability. Results also conclusively demonstrate the risks in normalizing data to creatinine.

Improving the Generalizability of Convolutional Neural Network-Based Segmentation on CMR Images.

Background: Convolutional neural network (CNN) based segmentation methods provide an efficient and automated way for clinicians to assess the structure and function of the heart in cardiac MR images. While CNNs can generally perform the segmentation tasks with high accuracy when training and test images come from the same domain (e.g., same scanner or site), their performance often degrades dramatically on images from different scanners or clinical sites.Methods: We propose a simple yet effective way for improving the network generalization ability by carefully designing data normalization and augmentation strategies to accommodate common scenarios in multi-site, multi-scanner clinical imaging data sets. We demonstrate that a neural network trained on a single-site single-scanner dataset from the UK Biobank can be successfully applied to segmenting cardiac MR images across different sites and different scanners without substantial loss of accuracy. Specifically, the method was trained on a large set of 3,975 subjects from the UK Biobank. It was then directly tested on 600 different subjects from the UK Biobank for intra-domain testing and two other sets for cross-domain testing: the ACDC dataset (100 subjects, 1 site, 2 scanners) and the BSCMR-AS dataset (599 subjects, 6 sites, 9 scanners).Results: The proposed method produces promising segmentation results on the UK Biobank test set which are comparable to previously reported values in the literature, while also performing well on cross-domain test sets, achieving a mean Dice metric of 0.90 for the left ventricle, 0.81 for the myocardium, and 0.82 for the right ventricle on the ACDC dataset; and 0.89 for the left ventricle, 0.83 for the myocardium on the BSCMR-AS dataset.Conclusions: The proposed method offers a potential solution to improve CNN-based model generalizability for the cross-scanner and cross-site cardiac MR image segmentation task.

Data-dependent normalization strategies for untargeted metabolomics-a case study.

Despite the recent advances in the standardization of untargeted metabolomics workflows, there is still a lack of attention to specific data treatment strategies that require deep knowledge of the biological problem and need to be applied after a well-thought out process to understand the effect of the practice. One of those strategies is data normalization. Data-driven assumptions are critical especially addressing unwanted variation present in the biological model as it can be the case in heterogeneous tissues, cells with different sizes or biofluids with different concentrations. Chronic kidney disease (CKD) is a widespread disorder affecting kidney structure and function. Animal models are being developed to be able to get valuable insights into the etiopathogenesis of the condition and effect of the treatments. Moreover, diagnosis and disease staging still require defining appropriate biomarkers. Untargeted metabolomics has the potential to deal with those challenges. Renal fibrosis is one of the consequences of kidney injury which greatly affects the concentration of metabolites in the same quantity of sample. To overcome this challenge, several data normalization strategies have been applied, following a multilevel normalization method with the overall aim of focussing on the relevant biological information and reducing the influence of disturbing factors. A comprehensive evaluation of the performance of the normalization strategies, both on methods assessing the intragroup variation and on the impact on differential analysis, is provided. Finally, we present evidence of the importance of biological-model-driven guided normalization methods and discuss multiple criteria that need to be taken into consideration to obtain robust and reliable data. Special concern is transmitted on the misleading conclusions that might be the consequence of inappropriate data pre-treatment solutions applied for untargeted methods. Graphical abstract.

Data normalization strategies in metabolomics: Current challenges, approaches, and tools.

Data normalization is a big challenge in quantitative metabolomics approaches, whether targeted or untargeted. Without proper normalization, the mass-spectrometry and spectroscopy data can provide erroneous, sub-optimal data, which can lead to misleading and confusing biological results and thereby result in failed application to human healthcare, clinical, and other research avenues. To address this issue, a number of statistical approaches and software tools have been proposed in the literature and implemented over the years, thereby providing a multitude of approaches to choose from - either sample-based or data-based normalization strategies. In recent years, new dedicated software tools for metabolomics data normalization have surfaced as well. In this account article, I summarize the existing approaches and the new discoveries and research findings in this area of metabolomics data normalization, and I introduce some recent tools that aid in data normalization.