Kevin R. Coombes

Deriving chemosensitivity from cell lines: Forensic bioinformatics and reproducible research in high-throughput biology

High-throughput biological assays such as microarrays let us ask very detailed questions about how diseases operate, and promise to let us personalize therapy. Data processing, however, is often not described well enough to allow for exact reproduction of the results, leading to exercises in “forensic bioinformatics” where aspects of raw data and reported results are used to infer what methods must have been employed. Unfortunately, poor documentation can shift from an inconvenience to an active danger when it obscures not just methods but errors. In this report, we examine several related papers purporting to use microarray-based signatures of drug sensitivity derived from cell lines to predict patient response. Patients in clinical trials are currently being allocated to treatment arms on the basis of these results. However, we show in five case studies that the results incorporate several simple errors that may be putting patients at risk. One theme that emerges is that the most common errors are simple (e.g., row or column offsets); conversely, it is our experience that the most simple errors are common. We then discuss steps we are taking to avoid such errors in our own investigations.

Functional proteomic profiling of AML predicts response and survival

Because protein function regulates the phenotypic characteristics of cancer, a functional proteomic classification system could provide important information for pathogenesis and prognosis. With the goal of ultimately developing a proteomic-based classification of acute myeloid leukemia (AML), we assayed leukemia-enriched cells from 256 newly diagnosed AML patients, for 51 total and phosphoproteins from apoptosis, cell-cycle, and signal-transduction pathways, using reverse-phase protein arrays. Expression in matched blood and marrow samples were similar for 44 proteins; another 7 had small fold changes (8%-55%), suggesting that functional proteomics of leukemia-enriched cells in the marrow and periphery are similar. Protein expression patterns were independent of clinical characteristics. However, 24 proteins were significantly different between French-American-British subtypes, defining distinct signatures for each. Expression signatures for AML with cytogenetic abnormalities involving -5 or -7 were similar suggesting mechanistic commonalities. Distinct expression patterns for FMS-like tyrosine kinase 3-internal tandem duplication were also identified. Principal component analysis defined 7 protein signature groups, with prognostic information distinct from cytogenetics that correlated with remission attainment, relapse, and overall survival. In conclusion, protein expression profiling patterns in AML correlate with known morphologic features, cytogenetics, and outcome. Confirmation in independent studies may also provide pathophysiologic insights facilitating triage of patients to emerging targeted therapies.

Applications of beta-mixture models in bioinformatics

SUMMARY We propose a beta-mixture model approach to solve a variety of problems related to correlations of gene-expression levels. For example, in meta-analyses of microarray gene-expression datasets, a threshold value of correlation coefficients for gene-expression levels is used to decide whether gene-expression levels are strongly correlated across studies. Ad hoc threshold values such as 0.5 are often used. In this paper, we use a beta-mixture model approach to divide the correlation coefficients into several populations so that the large correlation coefficients can be identified. Another important application of the proposed method is in finding co-expressed genes. Two examples are provided to illustrate both applications. Through our analysis, we also discover that the popular model selection criteria BIC and AIC are not suitable for the beta-mixture model. To determine the number of components in the mixture model, we suggest an alternative criterion, ICL-BIC, which is shown to perform better in selecting the correct mixture model. SUPPLEMENTARY INFORMATION http://odin.mdacc.tmc.edu/~yuanj/highcorgeneanno.html.

HOX expression patterns identify a common signature for favorable AML.

Deregulated HOX expression, by chromosomal translocations and myeloid-lymphoid leukemia (MLL) rearrangements, is causal in some types of leukemia. Using real-time reverse transcription-PCR, we examined the expression of 43 clustered HOX, polycomb, MLL and FLT3 genes in 119 newly diagnosed adult acute myeloid leukemias (AMLs) selected from all major cytogenetic groups. Downregulated HOX expression was a consistent feature of favorable AMLs and, among these cases, inv(16) cases had a distinct expression profile. Using a 17-gene predictor in 44 additional samples, we observed a 94.7% specificity for classifying favorable vs intermediate/unfavorable cytogenetic groups. Among other AMLs, HOX overexpression was associated with nucleophosmin (NPM) mutations and we also identified a phenotypically similar subset with wt-NPM. In many unfavorable and other intermediate cytogenetic AMLs, HOX levels resembled those in normal CD34+ cells, except that the homogeneity characteristic of normal samples was not present. We also observed that HOXA9 levels were significantly inversely correlated with survival and that BMI-1 was overexpressed in cases with 11q23 rearrangements, suggesting that p19ARF suppression may be involved in MLL-associated leukemia. These results underscore the close relationship between HOX expression patterns and certain forms of AML and emphasize the need to determine whether these differences play a role in the disease process.

Serial dilution curve

Reverse phase protein arrays (RPPAs) are a powerful high-throughput tool for measuring protein concentrations in a large number of samples. In RPPA technology, the original samples are often diluted successively multiple times, forming dilution series to extend the dynamic range of the measurements and to increase confidence in quantitation. An RPPA experiment is equivalent to running multiple ELISA assays concurrently except that there is usually no known protein concentration from which one can construct a standard response curve. Here, we describe a new method called ‘serial dilution curve for RPPA data analysis’. Compared with the existing methods, the new method has the advantage of using fewer parameters and offering a simple way of visualizing the raw data. We showed how the method can be used to examine data quality and to obtain robust quantification of protein concentrations. Availability: A computer program in R for using serial dilution curve for RPPA data analysis is freely available at http://odin.mdacc.tmc.edu/~zhangli/RPPA. Contact: lzhangli@mdanderson.org

PrepMS: TOF MS data graphical preprocessing tool

UNLABELLED We introduce a simple-to-use graphical tool that enables researchers to easily prepare time-of-flight mass spectrometry data for analysis. For ease of use, the graphical executable provides default parameter settings, experimentally determined to work well in most situations. These values, if desired, can be changed by the user. PrepMS is a stand-alone application made freely available (open source), and is under the General Public License (GPL). Its graphical user interface, default parameter settings, and display plots allow PrepMS to be used effectively for data preprocessing, peak detection and visual data quality assessment. AVAILABILITY Stand-alone executable files and Matlab toolbox are available for download at: http://sourceforge.net/projects/prepms

Reproducibility of SELDI Spectra Across Time and Laboratories

The reproducibility of mass spectrometry (MS) data collected using surface enhanced laser desorption/ionization-time of flight (SELDI-TOF) has been questioned. This investigation was designed to test the reproducibility of SELDI data collected over time by multiple users and instruments. Five laboratories prepared arrays once every week for six weeks. Spectra were collected on separate instruments in the individual laboratories. Additionally, all of the arrays produced each week were rescanned on a single instrument in one laboratory. Lab-to-lab and array-to-array variability in alignment parameters were larger than the variability attributable to running samples during different weeks. The coefficient of variance (CV) in spectrum intensity ranged from 25% at baseline, to 80% in the matrix noise region, to about 50% during the exponential drop from the maximum matrix noise. Before normalization, the median CV of the peak heights was 72% and reduced to about 20% after normalization. Additionally, for the spectra from a common instrument, the CV ranged from 5% at baseline, to 50% in the matrix noise region, to 20% during the drop from the maximum matrix noise. Normalization reduced the variability in peak heights to about 18%. With proper processing methods, SELDI instruments produce spectra containing large numbers of reproducibly located peaks, with consistent heights.

Inter-gene correlation on oligonucleotide arrays: how much does normalization matter?

AbstractBackground and objective: Normalization is a standard low-level preprocessing procedure in microarray data analysis to minimize the systematic technological variations and produce more reliable results. A variety of normalization approaches have been introduced and are widely applied. Normalization, however, remains controversial. The sensitivity of array results to normalization is an open question. No clear standard for comparing or judging normalization methods has yet emerged, and the effects of normalization on gene-to-gene co-expression are unclear. Methods: In this investigation, we applied 1-, 2-, and N-quantile normalization to several publicly available microarray datasets quantified with either MAS 5.0 or dCHIP and evaluated the effect on gene-to-gene co-expression. We introduced a graphical method to explore trends in gene correlation. Results: We found clear differences in the distributions of gene dependencies by normalization method. Increasing the number of standardized quantiles in the normalization reduced trends in correlation by signal intensity in MAS 5.0 quantifications but not dCHIP. Increasing the number of standardized quantiles did not markedly reduce the correlation of known overlapping targets with MAS 5.0. Conclusions: Normalization plays a very important role in the estimation of inter-gene dependency. Caution should be used when making inferences concerning gene-wise dependencies with microarrays until this source of variation is better understood.

Monitoring the Quality of Microarray Experiments

A microarray experiment is a complex, multistep process involving biology, chemistry, physics, and bioinformatics. Something can go wrong at every step in the process. In order to obtain good results, one needs a thorough, redundant system to monitor the quality of microarray experiments. In this article, we provide an overview of quality control measures that can be applied at different points during the process of conducting and analyzing microarray experiments.

Elliptic curves and logarithmic derivatives

Let C be a curve with Jacobian variety J defined over an arbitrary field k. In this paper, we show that the logarithmic derivative induces a natural homomorphism from the group J(k) of k-rational points on J into the group (H1(C,Oc) ⊗k ΩkZ1)δ(Γ(C, ΩCk 1)), where δ is a connecting homomorphism in a natural sequence of Zariski cohomology groups. When C = E is an elliptic curve with j-invariant equal to j, we show that the image of δ is the k-vector subspace of Ωkz1 spanned by the absolute differential dj. Thus, we can interpret the logarithmic derivative as a map dlog :E(k) → Ωk[j]z1. Finally, we compute the kernel of this morphism explicitly. To describe the main theorem, write the Weierstrass equation of E in the form y2 = x3 + a4x + a6. Let k0 be the prime field of k and let F be the algebraic closure in k of the field k0(a4, a6). We show that the kernel of dlog can be identified with the group E(F) of F-rational points on E. In particular, notice that when k = C is the field of complex numbers, then the kernel of dlog is countable, and its image must be uncountable.