Guy N. Brock

A comparison of per sample global scaling and per gene normalization methods for differential expression analysis of RNA-seq data

Normalization is an essential step with considerable impact on high-throughput RNA sequencing (RNA-seq) data analysis. Although there are numerous methods for read count normalization, it remains a challenge to choose an optimal method due to multiple factors contributing to read count variability that affects the overall sensitivity and specificity. In order to properly determine the most appropriate normalization methods, it is critical to compare the performance and shortcomings of a representative set of normalization routines based on different dataset characteristics. Therefore, we set out to evaluate the performance of the commonly used methods (DESeq, TMM-edgeR, FPKM-CuffDiff, TC, Med UQ and FQ) and two new methods we propose: Med-pgQ2 and UQ-pgQ2 (per-gene normalization after per-sample median or upper-quartile global scaling). Our per-gene normalization approach allows for comparisons between conditions based on similar count levels. Using the benchmark Microarray Quality Control Project (MAQC) and simulated datasets, we performed differential gene expression analysis to evaluate these methods. When evaluating MAQC2 with two replicates, we observed that Med-pgQ2 and UQ-pgQ2 achieved a slightly higher area under the Receiver Operating Characteristic Curve (AUC), a specificity rate > 85%, the detection power > 92% and an actual false discovery rate (FDR) under 0.06 given the nominal FDR (≤0.05). Although the top commonly used methods (DESeq and TMM-edgeR) yield a higher power (>93%) for MAQC2 data, they trade off with a reduced specificity (<70%) and a slightly higher actual FDR than our proposed methods. In addition, the results from an analysis based on the qualitative characteristics of sample distribution for MAQC2 and human breast cancer datasets show that only our gene-wise normalization methods corrected data skewed towards lower read counts. However, when we evaluated MAQC3 with less variation in five replicates, all methods performed similarly. Thus, our proposed Med-pgQ2 and UQ-pgQ2 methods perform slightly better for differential gene analysis of RNA-seq data skewed towards lowly expressed read counts with high variation by improving specificity while maintaining a good detection power with a control of the nominal FDR level.

Replication and validation of genetic polymorphisms associated with survival after allogeneic blood or marrow transplant

Multiple candidate gene-association studies of non-HLA single-nucleotide polymorphisms (SNPs) and outcomes after blood or marrow transplant (BMT) have been conducted. We identified 70 publications reporting 45 SNPs in 36 genes significantly associated with disease-related mortality, progression-free survival, transplant-related mortality, and/or overall survival after BMT. Replication and validation of these SNP associations were performed using DISCOVeRY-BMT (Determining the Influence of Susceptibility COnveying Variants Related to one-Year mortality after BMT), a well-powered genome-wide association study consisting of 2 cohorts, totaling 2888 BMT recipients with acute myeloid leukemia, acute lymphoblastic leukemia, or myelodysplastic syndrome, and their HLA-matched unrelated donors, reported to the Center for International Blood and Marrow Transplant Research. Gene-based tests were used to assess the aggregate effect of SNPs on outcome. None of the previously reported significant SNPs replicated at P < .05 in DISCOVeRY-BMT. Validation analyses showed association with one previously reported donor SNP at P < .05 and survival; more associations would be anticipated by chance alone. No gene-based tests were significant at P < .05. Functional annotation with publicly available data shows these candidate SNPs most likely do not have biochemical function; only 13% of candidate SNPs correlate with gene expression or are predicted to impact transcription factor binding. Of these, half do not impact the candidate gene of interest; the other half correlate with expression of multiple genes. These findings emphasize the peril of pursing candidate approaches and the importance of adequately powered tests of unbiased genome-wide associations with BMT clinical outcomes given the ultimate goal of improving patient outcomes.

Investigating skin-to-skin care patterns with extremely preterm infants in the NICU and their effect on early cognitive and communication performance: a retrospective cohort study

Objectives The primary objective of the study was to investigate how patterns of skin-to-skin care might impact infant early cognitive and communication performance. Design This was a retrospective cohort study. Setting This study took place in a level-IV all-referral neonatal intensive care unit in the Midwest USA specialising in the care of extremely preterm infants. Participants Data were collected from the electronic medical records of all extremely preterm infants (gestational age <27 weeks) admitted to the unit during 2010–2011 and who completed 6-month and 12-month developmental assessments in the follow-up clinic (n=97). Outcome measures Outcome measures included the cognitive and communication subscales of the Bayley Scales of Infant Development, Third Edition (Bayley-III); and skin-to-skin patterns including: total hours of maternal and paternal participation throughout hospitalisation, total duration in weeks and frequency (hours per week). Analysis Extracted data were analysed through a multistep process of logistic regressions, t-tests, χ2 tests and Fishers exact tests followed with exploratory network analysis using novel visual analytic software. Results Infants who received above the sample median in total hours, weekly frequency and total hours from mothers and fathers of skin-to-skin care were more likely to score ≥80 on the cognitive and communication scales of the Bayley-III. However, the results were not statistically significant (p>0.05). Mothers provided the majority of skin-to-skin care with a sharp decline at 30 weeks corrected age, regardless of when extremely preterm infants were admitted. Additional exploratory network analysis suggests that medical and skin-to-skin factors play a parallel, non-synergistic role in contributing to early cognitive and communication performance as assessed through the Bayley-III. Conclusions This study suggests an association between early and frequent skin-to-skin care with extremely preterm infants and early cognitive and communication performance.

Power analysis for RNA-Seq differential expression studies

BackgroundSample size calculation and power estimation are essential components of experimental designs in biomedical research. It is very challenging to estimate power for RNA-Seq differential expression under complex experimental designs. Moreover, the dependency among genes should be taken into account in order to obtain accurate results.ResultsIn this paper, we propose a simulation based procedure for power estimation using the negative binomial distribution and assuming a generalized linear model (at the gene level) that considers the dependence between gene expression level and its variance (dispersion) and also allows equal or unequal dispersion across conditions. We compared the performance of both Wald test and likelihood ratio test under different scenarios. The null distribution of the test statistics was simulated for the desired false positive control to avoid excess false positives with the usage of an asymptotic chi-square distribution. We applied this method to the TCGA breast cancer data set.ConclusionsWe provide a framework for power estimation of RNA-Seq data. The proposed procedure is able to properly control the false positive error rate at the nominal level.

Distribution based nearest neighbor imputation for truncated high dimensional data with applications to pre-clinical and clinical metabolomics studies

BackgroundHigh throughput metabolomics makes it possible to measure the relative abundances of numerous metabolites in biological samples, which is useful to many areas of biomedical research. However, missing values (MVs) in metabolomics datasets are common and can arise due to both technical and biological reasons. Typically, such MVs are substituted by a minimum value, which may lead to different results in downstream analyses.ResultsHere we present a modified version of the K-nearest neighbor (KNN) approach which accounts for truncation at the minimum value, i.e., KNN truncation (KNN-TN). We compare imputation results based on KNN-TN with results from other KNN approaches such as KNN based on correlation (KNN-CR) and KNN based on Euclidean distance (KNN-EU). Our approach assumes that the data follow a truncated normal distribution with the truncation point at the detection limit (LOD). The effectiveness of each approach was analyzed by the root mean square error (RMSE) measure as well as the metabolite list concordance index (MLCI) for influence on downstream statistical testing. Through extensive simulation studies and application to three real data sets, we show that KNN-TN has lower RMSE values compared to the other two KNN procedures as well as simpler imputation methods based on substituting missing values with the metabolite mean, zero values, or the LOD. MLCI values between KNN-TN and KNN-EU were roughly equivalent, and superior to the other four methods in most cases.ConclusionOur findings demonstrate that KNN-TN generally has improved performance in imputing the missing values of the different datasets compared to KNN-CR and KNN-EU when there is missingness due to missing at random combined with an LOD. The results shown in this study are in the field of metabolomics but this method could be applicable with any high throughput technology which has missing due to LOD.

Comparison of two equivalent model for end-stage liver disease scores for hepatocellular carcinoma patients using data from the United Network for Organ Sharing liver transplant waiting list registry

Patients with hepatocellular carcinoma (HCC) have been advantaged on the liver transplant waiting list within the United States, and a 6‐month delay and exception point cap have recently been implemented to address this disparity. An alternative approach to prioritization is an HCC‐specific scoring model such as the MELD Equivalent (MELDEQ) and the mixed new deMELD. Using data on adult patients added to the UNOS waitlist between 30 September 2009 and 30 June 2014, we compared projected dropout and transplant probabilities for patients with HCC under these two models. Both scores matched actual non‐HCC dropout in groups with scores <22 and improved equity with non‐HCC transplant probabilities overall. However, neither score matched non‐HCC dropout accurately for scores of 25–40 and projected dropout increased beyond non‐HCC probabilities for scores <16. The main differences between the two scores were as follows: (i) the MELDEQ assigns 6.85 more points after 6 months on the waitlist and (ii) the deMELD gives greater weight to tumor size and laboratory MELD. Post‐transplant survival was lower for patients with scores in the 22–30 range compared with those with scores <16 (P = 0.007, MELDEQ; P = 0.015, deMELD). While both scores result in better equity of waitlist outcomes compared with scheduled progression, continued development and calibration is recommended.

Mutational Mechanisms That Activate Wnt Signaling and Predict Outcomes in Colorectal Cancer Patients

APC biallelic loss-of-function mutations are the most prevalent genetic changes in colorectal tumors, but it is unknown whether these mutations phenocopy gain-of-function mutations in the CTNNB1 gene encoding β-catenin that also activate canonical WNT signaling. Here we demonstrate that these two mutational mechanisms are not equivalent. Furthermore, we show how differences in gene expression produced by these different mechanisms can stratify outcomes in more advanced human colorectal cancers. Gene expression profiling in Apc-mutant and Ctnnb1-mutant mouse colon adenomas identified candidate genes for subsequent evaluation of human TCGA (The Cancer Genome Atlas) data for colorectal cancer outcomes. Transcriptional patterns exhibited evidence of activated canonical Wnt signaling in both types of adenomas, with Apc-mutant adenomas also exhibiting unique changes in pathways related to proliferation, cytoskeletal organization, and apoptosis. Apc-mutant adenomas were characterized by increased expression of the glial nexin Serpine2, the human ortholog, which was increased in advanced human colorectal tumors. Our results support the hypothesis that APC-mutant colorectal tumors are transcriptionally distinct from APC-wild-type colorectal tumors with canonical WNT signaling activated by other mechanisms, with possible implications for stratification and prognosis.Significance: These findings suggest that colon adenomas driven by APC mutations are distinct from those driven by WNT gain-of-function mutations, with implications for identifying at-risk patients with advanced disease based on gene expression patterns. Cancer Res; 78(3); 617-30. ©2017 AACR.

Polychrome: Creating and Assessing Qualitative Palettes With Many Colors

Although R includes numerous tools for creating color palettes to display continuous data, facilities for displaying categorical data primarily use the RColorBrewer package, which is, by default, limited to 12 colors. The colorspace package can produce more colors, but it is not immediately clear how to use it to produce colors that can be reliably distingushed in different kinds of plots. However, applications to genomics would be enhanced by the ability to display at least the 24 human chromosomes in distinct colors, as is common in technologies like spectral karyotyping. In this article, we describe the Polychrome package, which can be used to construct palettes with at least 24 colors that can be distinguished by most people with normal color vision. Polychrome includes a variety of visualization methods allowing users to evaluate the proposed palettes. In addition, we review the history of attempts to construct qualitative color palettes with many colors.

An evaluation of veterinary allergen extract content and resultant canine intradermal threshold concentrations

BACKGROUND Limited information is available for dogs on threshold concentrations (TCs), and the protein composition of common allergenic extracts produced by different manufacturers. HYPOTHESIS/OBJECTIVES To characterize the protein heterogeneity of tree, grass, weed and mite allergens from different lots of allergenic extracts, and to determine intradermal TCs for healthy dogs using extracts from two manufacturers. ANIMALS Twenty five privately owned, clinically healthy dogs and ten purpose-bred beagle dogs. METHODS AND MATERIALS Protein concentration and heterogeneity of 11 allergens from two manufacturers were evaluated using a Bradford-style assay and SDS-PAGE. Intradermal testing was performed with 11 allergens from each company at four dilutions. Immediate reactions were subjectively scored (0 to 4+), and objectively measured (mm) and their percentage concordance evaluated. Model-based TCs were determined by fitting positive reactions (≥2+) at 15 min to generalized estimating equations. RESULTS Allergen extract protein quantity and composition varied within and between manufacturers despite sharing the same PNU/mL values. Model-based TCs of one weed, five trees, two grasses and a house dust mite were determined for extracts from Manufacturer 1 (M1), and for extracts of three weeds, three trees and two grasses from Manufacturer 2 (M2). Receiver operating characteristic curve analyses determined a percentage concordance of the objective and subjective measurements of 77.3% for M1 and 75% for M2 allergens. CONCLUSIONS AND CLINICAL IMPORTANCE Veterinary allergen extracts labelled as the same species and PNU/mL are not standardized; they show heterogeneity in composition and potency within and between manufacturers. Variability in extract content may require adjustment of intradermal testing concentrations.

A combined approach with gene-wise normalization improves the analysis of RNA-seq data in human breast cancer subtypes.

Breast cancer (BC) is increasing in incidence and resistance to treatment worldwide. The challenges in limited therapeutic options and poor survival outcomes in BC subtypes persist because of its molecular heterogeneity and resistance to standard endocrine therapy. Recently, high throughput RNA sequencing (RNA-seq) has been used to identify biomarkers of disease progression and signaling pathways that could be amenable to specific therapies according to the BC subtype. However, there is no single generally accepted pipeline for the analysis of RNA-seq data in biomarker discovery due, in part, to the needs of simultaneously satisfying constraints of sensitivity and specificity. We proposed a combined approach using gene-wise normalization, UQ-pgQ2, followed by a Wald test from DESeq2. Our approach improved the analysis based on within-group comparisons in terms of the specificity when applied to publicly available RNA-seq BC datasets. In terms of identifying differentially expressed genes (DEGs), we combined an optimized log2 fold change cutoff with a nominal false discovery rate of 0.05 to further minimize false positives. Using this method in the analysis of two GEO BC datasets, we identified 797 DEGs uniquely expressed in triple negative BC (TNBC) and significantly associated with T cell and immune-related signaling, contributing to the immunotherapeutic efficacy in TNBC patients. In contrast, we identified 1403 DEGs uniquely expressed in estrogen positive and HER2 negative BC (ER+HER2-BC) and significantly associated with eicosanoid, notching and FAK signaling while a common set of genes was associated with cellular growth and proliferation. Thus, our approach to control for false positives identified two distinct gene expression profiles associated with these two subtypes of BC which are distinguishable by their molecular and functional attributes.