Chris Wakefield

TCPA: a resource for cancer functional proteomics data

To the Editor: Functional proteomics represents a powerful approach to understand the pathophysiology and therapy of cancer. However, comprehensive cancer proteomic data have been relatively limited. As a part of The Cancer Genome Atlas (TCGA) Project and other efforts, we have generated protein expression data over a large number of tumor and cell line samples using reverse-phase protein arrays (RPPAs). RPPA is a quantitative, antibody-based technology that can assess multiple protein markers in many samples in a cost-effective, sensitive and highthroughput manner1,2. This technology has been extensively validated for both cell line and patient samples3–5, and its applications range from building reproducible prognostic models6 to generating experimentally verified mechanistic insights7. Our RPPA profiling platform includes extensively validated antibodies to nearly 200 proteins and phosphoproteins (Supplementary Methods and Supplementary Table 1). We are in the process of extending it to 500 independent proteins, covering all major signaling pathways, including PI3K, MAPK, mTOR, TGF-b, WNT, cell cycle, apoptosis, DNA damage, Hippo and Notch pathways. The current data release covers 4,379 tumor samples and consists of three parts (Supplementary Table 2). These are (i) TCGA tumor tissue sample sets: 3,467 samples from 11 cancer types, to be extended to 25 cancer types; (ii) independent tumor tissue sample sets: one endometrial tumor set (244 samples)7 and two ovarian tumor sets (99 and 130 samples, respectively)6, with other independent sets to be added soon; and (iii) tumor cell lines: 439 samples in four cell line sets, including both baseline and drug-treated cell lines. To our knowledge, this represents the largest publicly available collection of cancer functional proteomics data with parallel DNA and RNA data. To facilitate broad access to these RPPA data sets, we developed a user-friendly data portal, The Cancer Proteome Atlas (TCPA; http://bioinformatics.mdanderson.org/main/ TCPA:Overview). TCPA provides six modules: Summary, My Protein, Download, Visualization, Analysis and Cell Line (Fig. 1, i). The Summary module provides an overview of the RPPA data with detailed descriptions of each set (Fig. 1, ii). The Download module allows users to obtain any RPPA data set for analysis through a tree-view interface (Fig. 1, iii). The My Protein module provides detailed information about each RPPA protein: protein name, corresponding gene symbol, antibody status and source for the antibody. Users can examine the expression pattern of a protein of interest across different tumor types (for example, HER2 expression shown in Fig. 1, iv). The Visualization module provides two ways to examine global protein expression patterns in a specific RPPA data set . One is through a “next-generation clustered heat map” (Fig. 1, v), which allows users to zoom, navigate and scrutinize clustering patterns of samples or proteins and link those patterns to relevant biological information sources. The other is through a network view (Fig. 1, vi), which overlays the correlation between any two interacting partners in the protein interaction network (curated in the Human Protein Reference Database8). The Analysis module provides three analysis methods. (i) For correlation analysis, given a user-specified data set, correlations between any pair of proteins are presented in a table (Fig. 1, vii). Users can search the results by protein name, rank correlations or visualize the scatter plot of a correlation of interest (for example, there is a strong correlation between PKC-a and its phosphorylated form PKC-a_pS657 in endometrial cancer, as shown in Fig. 1, vii). (ii) For differential analysis, differentially expressed protein markers between two tumor types or subtypes can be identified. Given user-defined comparison groups, the Krzywinski and Cairo reply: We are in full agreement with the core of Katz’s argument that “distortion,” “embellishment,” “concealment” and “unrepresentative displays” have no place in principled communication of scientific information1. There is no controversy here—Katz extrapolates our storytelling metaphor beyond the intended scope of our column and argues against a position we did not take. The Points of View series offers effective strategies for visual presentation of complex data. The scope of the Storytelling column2 was limited to the construction of multipanel figures, which summarize as much as they support detailed exposition of the text. The column did not address how this text should be composed or the broad subject of motivation and design of scientific experiments. We described an approach to structure the flow of concepts and data across panels in a figure as a way to achieve a narrative, not confabulation. The design of visual communication requires a distinct approach because we organize and interpret images very differently than words (Gestalt principles of perception3). Whereas text is a natural place for nuance and alternative interpretations, multiple lines of argument in a figure can easily interfere with our perception of all its parts. Our suggestion to “leave out detail that does not advance the plot” speaks to controlling the amount of information to avoid an incomprehensible image and deferring it to the text, where it can be more suitably framed. To interpret it as “inconvenient truths are [to be] swept away” is a misrepresentation. Readers often look to the abstract and then the figures to provide them with an initial impression and overview of the findings. These are not the only elements that are reported, merely the first elements to be read. At each step, from abstract to figure to text, the level of detail is expanded to accommodate the preparedness of the reader to assimilate new information. It is often impossible to “do justice to experimental complexities and their myriad of interpretations” with a figure. We support Katz’s position that authors should include all the details necessary to appreciate, understand and reproduce the science through the use of visual and written communication that is clear, concise and thoughtful.

Automating and simplifying the SOFA score in critically ill patients with cancer

The aim was to demonstrate the performance of a modified version of the Sequential Organ Failure Assessment (SOFA) score to predict mortality in medical and surgical patients with cancer. We performed an electronic retrospective review of databases. We included adult patients with cancer admitted into a 53-bed ICU over 28 months. We electronically calculated a modified SOFA (mSOFA) score at admission. A majority of the patients were admitted into the surgical ICU. Of 328 nonsurvivors, 85.1 per cent were medical patients and only 14.9 per cent surgical patients. The mean admission mSOFA scores for medical and surgical patients were 4.7 ± 3.2 and 1.7 ± 1.9, respectively. The overall area under the curve (AUC) of the mSOFA score was 0.84. The AUCs for medical and surgical patients were 0.72 and 0.78, respectively. Our results demonstrate that electronic assessment of mSOFA score has potential in resource allocation decisions as well as in critical care outreach programs.

TransVar: a multilevel variant annotator for precision genomics

One DNA sequence can code for multiple different mRNAs, and therefore many different proteins. Conversely, a variant identified at the protein or transcript level may have non-unique genomic origins. For example, EGFR:p.L747S, which mediates acquired resistance of non-small cell lung cancer to tyrosine kinase inhibitors1, can be translated from multiple genomic variants such as chr7:g.55249076_55249077delinsAG and chr7:g.55242470T>C on different isoforms defined on the human reference assembly GRCh37. One-to-many, many-to-one and many-to-many relationships among sequence variants at the genomic level and those at transcript and protein levels introduce frequent inconsistencies in current practice when vital information about the annotation process (e.g., transcript or isoform IDs) is omitted from variant identifiers. To facilitate standardization and reveal inconsistency in existing variant annotations, we have designed a novel variant annotator, TransVar, to perform three main functions supporting diverse reference genomes and transcript databases (Fig. 1a): (i) “forward annotation”, which annotates all potential effects of a genomic variant on mRNAs and proteins; (ii) “reverse annotation”, which traces an mRNA or protein variant to all potential genomic origins; and (iii) “equivalence annotation”, which, for a given protein variant, searches for alternative protein variants that have identical genomic origin but are represented based on different isoforms. Figure 1 Schematic overview of TransVar and comparison of TransVar with other tools. (a) TransVar performs forward (green arrows) and reverse annotation (pink arrows) and considers all possible mRNA transcripts or protein isoforms available in user-specified reference ... We annotated 964,132 unique single-nucleotide substitutions (SNS), 3,715 multi-nucleotide substitutions (MNS), 11,761 insertions (INS), 24,595 deletions (DEL) and 166 block substitutions (BLS) in the Catalogue of Somatic Mutations in Cancer (COSMIC v67) using TransVar, ANNOVAR2, VEP3, snpEff4, and Oncotator5, and asked whether the resulting protein identifiers (gene name, protein coordinates, and reference amino acid (AA)) match those in COSMIC. We observed comparable consistency in SNS and MNS but variable consistency in INS, DEL and BLS from different annotators (Fig. 1b, Supplementary Table 1 and Supplementary Notes). That finding can largely be attributed to a lack of standardization among variant annotations (codon or AA positions of variants) submitted to COSMIC and among conventions implemented in various annotators. Inconsistency in annotations blurred the lines of evidence for variant frequency estimation and led to inaccurate determination of variant function. TransVar revealed hidden inconsistency in these variant annotations by comprehensively outputting alternative annotations in all available transcripts in standard HGVS nomenclature, and thus resulted in greater consistency in this experiment. TransVar’s novel reverse annotation can be used to ascertain if two protein variants have identical genomic origin, thus reducing inconsistency in annotation data. It can also reveal whether or not a protein variant has non-unique genomic origins and requires caution in genetic and clinical interpretation. We reverse-annotated the protein level variants in COSMIC and found that even under the constraints imposed by the reference base or AA identity, a sizeable fraction (e.g., 11.9% of single-AA substitutions) were associated with multiple genomic variants (Supplementary Table 2), if transcripts were not specified. Among the 537 variants that were cited as clinically actionable at PersonalizedCancerTherapy.org, 78 (14.5%) (e.g., CDKN2A:p.R87P and ERBB2:p.L755_T759del) could be mapped to multiple genomic locations (Supplementary Table 3). The reverse-annotation functionality also enabled systematic genomic characterization of variants directly identified from proteomic or RNA-seq data. For example, we were able to identify in just a few minutes of compute-time the putative genomic origins of 187,464 (97.69%) protein phosphorylation sites (e.g., p.Y308/p.S473 in AKT1 and p.Y1068/p.Y1172 in EGFR) in human proteins6. Our investigation revealed frequent inconsistencies in current databases and tools and highlighted the importance of standardization. With both forward and reverse annotation enabled in TransVar, we can reveal hidden inconsistency and improve the precision of translational and clinical genomics. The source code and detailed instructions of TransVar is available at https://bitbucket.org/wanding/transvar and a web interface is at http://www.transvar.net.

Cross-validation of a Sequential Organ Failure Assessment score–based model to predict mortality in patients with cancer admitted to the intensive care unit☆☆☆

PURPOSE This study aims to validate the performance of the Sequential Organ Failure Assessment (SOFA) score to predict death of critically ill patients with cancer. MATERIAL AND METHODS We conducted a retrospective observational study including adults admitted to the intensive care unit (ICU) between January 1, 2006, and December 31, 2008. We randomly selected training and validation samples in medical and surgical admissions to predict ICU and in-hospital mortality. By using logistic regression, we calculated the probabilities of death in the training samples and applied them to the validation samples to test the goodness-of-fit of the models, construct receiver operator characteristics curves, and calculate the areas under the curve (AUCs). RESULTS In predicting mortality at discharge from the unit, the AUC from the validation group of medical admissions was 0.7851 (95% confidence interval [CI], 0.7437-0.8264), and the AUC from the surgical admissions was 0.7847 (95% CI, 0.6319-0.937). The AUCs of the SOFA score to predict mortality in the hospital after ICU admission were 0.7789 (95% CI, 0.74-0.8177) and 0.7572 (95% CI, 0.6719-0.8424) for the medical and surgical validations groups, respectively. CONCLUSIONS The SOFA score had good discrimination to predict ICU and hospital mortality. However, the observed underestimation of ICU deaths and unsatisfactory goodness-of-fit test of the model in surgical patients to indicate calibration of the score to predict ICU mortality is advised in this group.

Life-Supportive Therapy Withdrawal and Length of Stay in a Large Oncologic Intensive Care Unit at the End of Life

We evaluated the factors associated with life-supportive therapy withdrawal (LSTW) and length-of-stay (LOS) of adult patients with cancer who died while in the intensive care unit (ICU). We performed chart review of adult patients with cancer who died in a 53-bed ICU of a comprehensive cancer center and evaluated the relative impact of demographic and clinical factors by using logistic regression and linear regression. A total of 267 patients were included in the study. Multivariate analysis showed that white patients were 2.52 times more likely to have LSTW than patients of other ethnicities (95% confidence interval 1.23-5.15, p = 0.01). The mean LOS for patients with hematologic cancers was 7.7 days, compared to only 4.8 days for those diagnosed with nonhematologic cancers (p = 0.01). Having a hematologic cancer, LSTW, or admission into the surgical oncologic ICU independently predicted increased LOS for those who died in the ICU (p < 0.001 p = 0.001, and p < 0.001, respectively). Cultural differences in dealing with the end-of-life process rooted in religious beliefs or language barriers and reflected in the utilization rates of LSTW by non-whites and whites may partially explain our findings. The difficult transition from curative to palliative care in the ICU is reflected by the increased LOS of patients who received LSTW, were diagnosed with hematologic cancers, or were admitted into the surgical unit.

Stress-related mucosal bleeding in critically ill oncology patients

Objective. To determine the incidence of stress-related mucosal bleeding (SRMB) in a critically ill oncology population receiving stress ulcer prophylaxis (SUP) with either a histamine-2 receptor antagonist (H2RA) or proton pump inhibitor (PPI). Design. Single-center, prospective, observational study. Setting. Fifty-two bed medical-surgical intensive care unit of an academic oncology institution. Patients. A convenience sample of 100 medical and surgical critically ill oncology patients who received intensive care for more than 24 hours and at least one dose of a H2RA or PPI for prevention of SRMB. Interventions. None. Measurements and Main Results. Patients were followed throughout their intensive care unit stay for the development of an overt and/or clinically significant gastrointestinal (GI) bleed. More patients received a PPI (n = 81) in contrast to a H2RA (n = 19) for SUP. Overall, 94 patients (94%) had at least one risk factor for a SRMB with four patients (4%) experiencing an event (overt bleed, n=3; clinically significant bleed, n =1). All cases of GI bleeding occurred in patients receiving a PPI. No ICU deaths were considered directly related to a GI bleed. Conclusions. The incidence of SRMB among high-risk critically ill oncology patients receiving SUP appears low; further, large-scale trials are needed to confirm this finding. J Oncol Pharm Practice (2009) 15: 9—16.

Factors associated with anemia in patients with cancer admitted to an intensive care unit

PURPOSE The study aimed to evaluate the relative impact of clinical and demographic factors associated with the prevalence and incidence of anemia (hemoglobin [Hb] <12 g/dL) in critically ill patients with cancer. MATERIALS AND METHODS We performed an electronic chart review for demographic and clinical data of adult patients with cancer with or without anemia admitted to the intensive care unit (ICU). Prevalence of anemia was determined at admission, and incidence determined if anemia developed during ICU stay. Anemia was classified as mild, moderate, or severe. The additive impact of clinical and demographic factors was evaluated by using a hierarchical linear regression model. RESULTS A total of 4705 patients were included in the study. The prevalence and incidence of anemia were 68.0% and 46.6%, respectively. In prevalent cases, we found that the clinical covariates modified sequential organ failure assessment score, admission to the medical ICU, prior chemotherapy, diagnosis of hematologic cancer, and length of hospital stay before ICU admission explained 18.7% of the variance in the model, whereas the demographic covariates (age, sex, and race) explained only an additional 0.6%. The pattern was similar for incidence cases. CONCLUSIONS Clinical factors are more influential than demographic factors in the observed rates of prevalence and incidence of anemia in the ICU; thus, protocols are needed to identify subgroups of patients with cancer who could benefit from novel management strategies.

SoS Notebook: an interactive multi-language data analysis environment

Motivation: Complex bioinformatic data analysis workflows involving multiple scripts in different languages can be difficult to consolidate, share and reproduce. An environment that streamlines the entire processes of data collection, analysis, visualization and reporting of such multi‐language analyses is currently lacking. Results: We developed Script of Scripts (SoS) Notebook, a web‐based notebook environment that allows the use of multiple scripting language in a single notebook, with data flowing freely within and across languages. SoS Notebook enables researchers to perform sophisticated bioinformatic analysis using the most suitable tools for different parts of the workflow, without the limitations of a particular language or complications of cross‐language communications. Availability and implementation: SoS Notebook is hosted at http://vatlab.github.io/SoS/ and is distributed under a BSD license.

Abstract 5132: Interactively exploring patterns in TCGA data: a web-based compendium of ‘next-generation’ clustered heat maps.

Each of the 5 TCGA marker paper published in Nature to date has included at least one clustered heat map (CHM). We introduced CHMs in the early 1990’s for pharmacogenomic analysis (1) and later for integrated visualization of genomic, transcriptomic, proteomic, pharmacological, and functional data (1). As the ubiquitous first-order way of visualizing omic data, CHMs have appeared in many thousands of publications (3–9), including those from TCGA. We have elsewhere summarized their limitations (10). One such limitation is that CHMs are generally static images. We therefore initiated the next-generation CHM (NG-CHM) project, using an image-tiling technology similar to that in Google Maps for navigation and extreme drill-down without loss of resolution. Once the CHM has been zoomed sufficiently, labels (e.g., gene, protein, or drug names) appear on the image9s axes. Clicking on a label produces a menu of link-outs (e.g., to GeneCards, Google, PubMed). For gene vs. gene maps, each pixel can represent a color-coded Pearson correlation coefficient. Clicking on the pixel pulls up the corresponding data scattergram, bootstrap statistics, literature references, or pathway relationships. Strong usability features include floating windows, flexible search tools, cluster selection tools, customizable re-coloring of the CHM, and high-quality PDF9s suitable for publication. NG-CHMs are a major resource for exploratory analysis and visualization in multiple projects of TCGA and other large-scale molecular profiling programs. Explore interactive versions for TCGA breast, colorectal, lung squamous, and glioblastoma data at http://bioinformatics.mdanderson.org/main/TCGA/NGCHM. Supported in part by NCI Grant No. U24CA143883, by a gift from the Mary K. Chapman Foundation, and by a grant from the Michael and Susan Dell Foundation honoring Lorraine Dell. Citation Format: John N. Weinstein, David W. Kane, Rehan Akbani, Deepti Dodda, Lam Nguyen, Michael C. Ryan, Chris Wakefield, Bradley M. Broom. Interactively exploring patterns in TCGA data: a web-based compendium of ‘next-generation’ clustered heat maps. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 5132. doi:10.1158/1538-7445.AM2013-5132