Bayesian Network structure learning algorithm for highly missing and non imputable data: Application to breast cancer radiotherapy data.

It is not uncommon for real-life data produced in healthcare to have a higher proportion of missing data than in other scopes. To take into account these missing data, imputation is not always desired by healthcare experts, and complete case analysis can lead to a significant loss of data. The algorithm proposed here, allows the learning of Bayesian Network graphs when both imputation and complete case analysis are not possible. The learning process is based on a set of local bootstrap learnings performed on complete sub-datasets which are then aggregated and locally optimized. This learning method presents competitive results compared to other structure learning algorithms, whatever the mechanism of missing data.

Towards a More Accurate Differential Analysis of Multiple Imputed Proteomics Data with mi4limma.

Imputing missing values is a common practice in label-free quantitative proteomics. Imputation replaces a missing value by a user-defined one. However, the imputation itself is not optimally considered downstream of the imputation process. In particular, imputed datasets are considered as if they had always been complete. The uncertainty due to the imputation is not properly taken into account. Hence, the mi4p package provides a more accurate statistical analysis of multiple-imputed datasets. A rigorous multiple imputation methodology is implemented, leading to a less biased estimation of parameters and their variability, thanks to Rubin's rules. The imputation-based peptide's intensities' variance estimator is then moderated using Bayesian hierarchical models. This estimator is finally included in moderated t-test statistics to provide differential analyses results.

How visual attention span and phonological skills contribute to N170 print tuning: An EEG study in French dyslexic students.

Developmental dyslexia is a disorder characterized by a sustainable learning deficit in reading. Based on ERP-driven approaches focusing on the visual word form area, electrophysiological studies have pointed a lack of visual expertise for written word recognition in dyslexic readers by contrasting the left-lateralized N170 amplitudes elicited by alphabetic versus non-alphabetic stimuli. Here, we investigated in 22 dyslexic participants and 22 age-matched control subjects how two behavioural abilities potentially affected in dyslexic readers (phonological and visual attention skills) contributed to the N170 expertise during a word detection task. Consistent with literature, dyslexic participants exhibited poorer performance in these both abilities as compared to healthy subjects. At the brain level, we observed (1) an unexpected preservation of the N170 expertise in the dyslexic group suggesting a possible compensatory mechanism and (2) a modulation of this expertise only by phonological skills, providing evidence for the phonological mapping deficit hypothesis.

Accounting for multiple imputation-induced variability for differential analysis in mass spectrometry-based label-free quantitative proteomics.

Imputing missing values is common practice in label-free quantitative proteomics. Imputation aims at replacing a missing value with a user-defined one. However, the imputation itself may not be optimally considered downstream of the imputation process, as imputed datasets are often considered as if they had always been complete. Hence, the uncertainty due to the imputation is not adequately taken into account. We provide a rigorous multiple imputation strategy, leading to a less biased estimation of the parameters' variability thanks to Rubin's rules. The imputation-based peptide's intensities' variance estimator is then moderated using Bayesian hierarchical models. This estimator is finally included in moderated t-test statistics to provide differential analyses results. This workflow can be used both at peptide and protein-level in quantification datasets. Indeed, an aggregation step is included for protein-level results based on peptide-level quantification data. Our methodology, named mi4p, was compared to the state-of-the-art limma workflow implemented in the DAPAR R package, both on simulated and real datasets. We observed a trade-off between sensitivity and specificity, while the overall performance of mi4p outperforms DAPAR in terms of F-Score.

The MHC class I MICA gene is a histocompatibility antigen in kidney transplantation.

The identity of histocompatibility loci, besides human leukocyte antigen (HLA), remains elusive. The major histocompatibility complex (MHC) class I MICA gene is a candidate histocompatibility locus. Here, we investigate its role in a French multicenter cohort of 1,356 kidney transplants. MICA mismatches were associated with decreased graft survival (hazard ratio (HR), 2.12; 95% confidence interval (CI): 1.45-3.11; P < 0.001). Both before and after transplantation anti-MICA donor-specific antibodies (DSA) were strongly associated with increased antibody-mediated rejection (ABMR) (HR, 3.79; 95% CI: 1.94-7.39; P < 0.001; HR, 9.92; 95% CI: 7.43-13.20; P < 0.001, respectively). This effect was synergetic with that of anti-HLA DSA before and after transplantation (HR, 25.68; 95% CI: 3.31-199.41; P = 0.002; HR, 82.67; 95% CI: 33.67-202.97; P < 0.001, respectively). De novo-developed anti-MICA DSA were the most harmful because they were also associated with reduced graft survival (HR, 1.29; 95% CI: 1.05-1.58; P = 0.014). Finally, the damaging effect of anti-MICA DSA on graft survival was confirmed in an independent cohort of 168 patients with ABMR (HR, 1.71; 95% CI: 1.02-2.86; P = 0.041). In conclusion, assessment of MICA matching and immunization for the identification of patients at high risk for transplant rejection and loss is warranted.

Fitting and Cross-Validating Cox Models to Censored Big Data With Missing Values Using Extensions of Partial Least Squares Regression Models.

Fitting Cox models in a big data context -on a massive scale in terms of volume, intensity, and complexity exceeding the capacity of usual analytic tools-is often challenging. If some data are missing, it is even more difficult. We proposed algorithms that were able to fit Cox models in high dimensional settings using extensions of partial least squares regression to the Cox models. Some of them were able to cope with missing data. We were recently able to extend our most recent algorithms to big data, thus allowing to fit Cox model for big data with missing values. When cross-validating standard or extended Cox models, the commonly used criterion is the cross-validated partial loglikelihood using a naive or a van Houwelingen scheme -to make efficient use of the death times of the left out data in relation to the death times of all the data. Quite astonishingly, we will show, using a strong simulation study involving three different data simulation algorithms, that these two cross-validation methods fail with the extensions, either straightforward or more involved ones, of partial least squares regression to the Cox model. This is quite an interesting result for at least two reasons. Firstly, several nice features of PLS based models, including regularization, interpretability of the components, missing data support, data visualization thanks to biplots of individuals and variables -and even parsimony or group parsimony for Sparse partial least squares or sparse group SPLS based models, account for a common use of these extensions by statisticians who usually select their hyperparameters using cross-validation. Secondly, they are almost always featured in benchmarking studies to assess the performance of a new estimation technique used in a high dimensional or big data context and often show poor statistical properties. We carried out a vast simulation study to evaluate more than a dozen of potential cross-validation criteria, either AUC or prediction error based. Several of them lead to the selection of a reasonable number of components. Using these newly found cross-validation criteria to fit extensions of partial least squares regression to the Cox model, we performed a benchmark reanalysis that showed enhanced performances of these techniques. In addition, we proposed sparse group extensions of our algorithms and defined a new robust measure based on the Schmid score and the R coefficient of determination for least absolute deviation: the integrated R Schmid Score weighted. The R-package used in this article is available on the CRAN, http://cran.r-project.org/web/packages/plsRcox/index.html. The R package bigPLS will soon be available on the CRAN and, until then, is available on Github https://github.com/fbertran/bigPLS.

Temporal multiomic modeling reveals a B-cell receptor proliferative program in chronic lymphocytic leukemia.

B-cell receptor (BCR) signaling is crucial for the pathophysiology of most mature B-cell lymphomas/leukemias and has emerged as a therapeutic target whose effectiveness remains limited by the occurrence of mutations. Therefore, deciphering the cellular program activated downstream this pathway has become of paramount importance for the development of innovative therapies. Using an original ex vivo model of BCR-induced proliferation of chronic lymphocytic leukemia cells, we generated 108 temporal transcriptional and proteomic profiles from 1 h up to 4 days after BCR activation. This dataset revealed a structured temporal response composed of 13,065 transcripts and 4027 proteins, comprising a leukemic proliferative signature consisting of 430 genes and 374 proteins. Mathematical modeling of this complex cellular response further highlighted a transcriptional network driven by 14 early genes linked to proteins involved in cell proliferation. This group includes expected genes (EGR1/2, NF-kB) and genes involved in NF-kB signaling modulation (TANK, ROHF) and immune evasion (KMO, IL4I1) that have not yet been associated with leukemic cells proliferation. Our study unveils the BCR-activated proliferative genetic program in primary leukemic cells. This approach combining temporal measurements with modeling allows identifying new putative targets for innovative therapy of lymphoid malignancies and also cancers dependent on ligand-receptor interactions.

Combining label-free and label-based accurate quantifications with SWATH-MS: Comparison with SRM and PRM for the evaluation of bovine muscle type effects.

Mass spectrometry has proven to be a valuable tool for the accurate quantification of proteins. In this study, the performances of three targeted approaches, namely selected reaction monitoring (SRM), parallel reaction monitoring (PRM) and sequential windowed acquisition of all theoretical fragment ion mass spectra (SWATH-MS), to accurately quantify ten potential biomarkers of beef meat tenderness or marbling in a cohort of 64 muscle samples were evaluated. So as to get the most benefit out of the complete MS2 maps that are acquired in SWATH-MS, an original label-free quantification method to estimate protein amounts using an I-spline regression model was developed. Overall, SWATH-MS outperformed SRM in terms of sensitivity and dynamic range, while PRM still performed the best, and all three strategies showed similar quantification accuracies and precisions for the absolute quantification of targets of interest. This targeted picture was extended by 585 additional proteins for which amounts were estimated using the label-free approach on SWATH-MS; thus, offering a more global profiling of muscle proteomes and further insights into muscle type effect on candidate biomarkers of beef meat qualities as well as muscle metabolism.

selectBoost: a general algorithm to enhance the performance of variable selection methods.

MOTIVATION: With the growth of big data, variable selection has become one of the critical challenges in statistics. Although many methods have been proposed in the literature, their performance in terms of recall (sensitivity) and precision (predictive positive value) is limited in a context where the number of variables by far exceeds the number of observations or in a highly correlated setting. RESULTS: In this article, we propose a general algorithm, which improves the precision of any existing variable selection method. This algorithm is based on highly intensive simulations and takes into account the correlation structure of the data. Our algorithm can either produce a confidence index for variable selection or be used in an experimental design planning perspective. We demonstrate the performance of our algorithm on both simulated and real data. We then apply it in two different ways to improve biological network reverse-engineering. AVAILABILITY AND IMPLEMENTATION: Code is available as the SelectBoost package on the CRAN, https://cran.r-project.org/package=SelectBoost. Some network reverse-engineering functionalities are available in the Patterns CRAN package, https://cran.r-project.org/package=Patterns. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Compatibility at amino acid position 98 of MICB reduces the incidence of graft-versus-host disease in conjunction with the CMV status.

Graft-versus-host disease (GVHD) and cytomegalovirus (CMV)-related complications are leading causes of mortality after unrelated-donor hematopoietic cell transplantation (UD-HCT). The non-conventional MHC class I gene MICB, alike MICA, encodes a stress-induced polymorphic NKG2D ligand. However, unlike MICA, MICB interacts with the CMV-encoded UL16, which sequestrates MICB intracellularly, leading to immune evasion. Here, we retrospectively analyzed the impact of mismatches in MICB amino acid position 98 (MICB98), a key polymorphic residue involved in UL16 binding, in 943 UD-HCT pairs who were allele-matched at HLA-A, -B, -C, -DRB1, -DQB1 and MICA loci. HLA-DP typing was further available. MICB98 mismatches were significantly associated with an increased incidence of acute (grade II-IV: HR, 1.20; 95% CI, 1.15 to 1.24; P < 0.001; grade III-IV: HR, 2.28; 95% CI, 1.56 to 3.34; P < 0.001) and chronic GVHD (HR, 1.21; 95% CI, 1.10 to 1.33; P < 0.001). MICB98 matching significantly reduced the effect of CMV status on overall mortality from a hazard ratio of 1.77 to 1.16. MICB98 mismatches showed a GVHD-independent association with a higher incidence of CMV infection/reactivation (HR, 1.84; 95% CI, 1.34 to 2.51; P < 0.001). Hence selecting a MICB98-matched donor significantly reduces the GVHD incidence and lowers the impact of CMV status on overall survival.

Determining the number of components in PLS regression on incomplete data set.

Partial least squares regression - or PLS regression - is a multivariate method in which the model parameters are estimated using either the SIMPLS or NIPALS algorithm. PLS regression has been extensively used in applied research because of its effectiveness in analyzing relationships between an outcome and one or several components. Note that the NIPALS algorithm can provide estimates parameters on incomplete data. The selection of the number of components used to build a representative model in PLS regression is a central issue. However, how to deal with missing data when using PLS regression remains a matter of debate. Several approaches have been proposed in the literature, including the Q2 criterion, and the AIC and BIC criteria. Here we study the behavior of the NIPALS algorithm when used to fit a PLS regression for various proportions of missing data and different types of missingness. We compare criteria to select the number of components for a PLS regression on incomplete data set and on imputed data set using three imputation methods: multiple imputation by chained equations, k-nearest neighbour imputation, and singular value decomposition imputation. We tested various criteria with different proportions of missing data (ranging from 5% to 50%) under different missingness assumptions. Q2-leave-one-out component selection methods gave more reliable results than AIC and BIC-based ones.

Multi-omics dataset to decipher the complexity of drug resistance in diffuse large B-cell lymphoma.

The prognosis of patients with relapsed/refractory (R/R) diffuse large B-cell lymphoma (DLBCL) remains unsatisfactory and, despite major advances in genomic studies, the biological mechanisms underlying chemoresistance are still poorly understood. We conducted for the first time a large-scale differential multi-omics investigation on DLBCL patient's samples in order to identify new biomarkers that could early identify patients at risk of R/R disease and to identify new targets that could determine chemorefractoriness. We compared a well-characterized cohort of R/R versus chemosensitive DLBCL patients by combining label-free quantitative proteomics and targeted RNA sequencing performed on the same tissues samples. The cross-section of both data levels allowed extracting a sub-list of 22 transcripts/proteins pairs whose expression levels significantly differed between the two groups of patients. In particular, we identified significant targets related to tumor metabolism (Hexokinase 3), microenvironment (IDO1, CXCL13), cancer cells proliferation, migration and invasion (S100 proteins) or BCR signaling pathway (CD79B). Overall, this study revealed several extremely promising biomarker candidates related to DLBCL chemorefractoriness and highlighted some new potential therapeutic drug targets. The complete datasets have been made publically available and should constitute a valuable resource for the future research.

BCR-associated factors driving chronic lymphocytic leukemia cells proliferation ex vivo.

A chronic antigenic stimulation is believed to sustain the leukemogenic development of chronic lymphocytic leukemia (CLL) and most of lymphoproliferative malignancies developed from mature B cells. Reproducing a proliferative stimulation ex vivo is critical to decipher the mechanisms of leukemogenesis in these malignancies. However, functional studies of CLL cells remains limited since current ex vivo B cell receptor (BCR) stimulation protocols are not sufficient to induce the proliferation of these cells, pointing out the need of mandatory BCR co-factors in this process. Here, we investigated benefits of several BCR co-stimulatory molecules (IL-2, IL-4, IL-15, IL-21 and CD40 ligand) in multiple culture conditions. Our results demonstrated that BCR engagement (anti-IgM ligation) concomitant to CD40 ligand, IL-4 and IL-21 stimulation allowed CLL cells proliferation ex vivo. In addition, we established a proliferative advantage for ZAP70 positive CLL cells, associated to an increased phosphorylation of ZAP70/SYK and STAT6. Moreover, the use of a tri-dimensional matrix of methylcellulose and the addition of TLR9 agonists further increased this proliferative response. This ex vivo model of BCR stimulation with T-derived cytokines is a relevant and efficient model for functional studies of CLL as well as lymphoproliferative malignancies.

Matching for the nonconventional MHC-I MICA gene significantly reduces the incidence of acute and chronic GVHD.

Graft-versus-host disease (GVHD) is among the most challenging complications in unrelated donor hematopoietic cell transplantation (HCT). The highly polymorphic MHC class I chain-related gene A, MICA, encodes a stress-induced glycoprotein expressed primarily on epithelia. MICA interacts with the invariant activating receptor NKG2D, expressed by cytotoxic lymphocytes, and is located in the MHC, next to HLA-B Hence, MICA has the requisite attributes of a bona fide transplantation antigen. Using high-resolution sequence-based genotyping of MICA, we retrospectively analyzed the clinical effect of MICA mismatches in a multicenter cohort of 922 unrelated donor HLA-A, HLA-B, HLA-C, HLA-DRB1, and HLA-DQB1 10/10 allele-matched HCT pairs. Among the 922 pairs, 113 (12.3%) were mismatched in MICA MICA mismatches were significantly associated with an increased incidence of grade III-IV acute GVHD (hazard ratio [HR], 1.83; 95% confidence interval [CI], 1.50-2.23; P < .001), chronic GVHD (HR, 1.50; 95% CI, 1.45-1.55; P < .001), and nonelapse mortality (HR, 1.35; 95% CI, 1.24-1.46; P < .001). The increased risk for GVHD was mirrored by a lower risk for relapse (HR, 0.50; 95% CI, 0.43-0.59; P < .001), indicating a possible graft-versus-leukemia effect. In conclusion, when possible, selecting a MICA-matched donor significantly influences key clinical outcomes of HCT in which a marked reduction of GVHD is paramount. The tight linkage disequilibrium between MICA and HLA-B renders identifying a MICA-matched donor readily feasible in clinical practice.

Deviance residuals-based sparse PLS and sparse kernel PLS regression for censored data.

MOTIVATION: A vast literature from the past decade is devoted to relating gene profiles and subject survival or time to cancer recurrence. Biomarker discovery from high-dimensional data, such as transcriptomic or single nucleotide polymorphism profiles, is a major challenge in the search for more precise diagnoses. The proportional hazard regression model suggested by Cox (1972), to study the relationship between the time to event and a set of covariates in the presence of censoring is the most commonly used model for the analysis of survival data. However, like multivariate regression, it supposes that more observations than variables, complete data, and not strongly correlated variables are available. In practice, when dealing with high-dimensional data, these constraints are crippling. Collinearity gives rise to issues of over-fitting and model misidentification. Variable selection can improve the estimation accuracy by effectively identifying the subset of relevant predictors and enhance the model interpretability with parsimonious representation. To deal with both collinearity and variable selection issues, many methods based on least absolute shrinkage and selection operator penalized Cox proportional hazards have been proposed since the reference paper of Tibshirani. Regularization could also be performed using dimension reduction as is the case with partial least squares (PLS) regression. We propose two original algorithms named sPLSDR and its non-linear kernel counterpart DKsPLSDR, by using sparse PLS regression (sPLS) based on deviance residuals. We compared their predicting performance with state-of-the-art algorithms on both simulated and real reference benchmark datasets. RESULTS: sPLSDR and DKsPLSDR compare favorably with other methods in their computational time, prediction and selectivity, as indicated by results based on benchmark datasets. Moreover, in the framework of PLS regression, they feature other useful tools, including biplots representation, or the ability to deal with missing data. Therefore, we view them as a useful addition to the toolbox of estimation and prediction methods for the widely used Cox's model in the high-dimensional and low-sample size settings. AVAILABILITY AND IMPLEMENTATION: The R-package plsRcox is available on the CRAN and is maintained by Frédéric Bertrand. http://cran.r-project.org/web/packages/plsRcox/index.html. CONTACT: pbastien@rd.loreal.com or fbertran@math.unistra.fr. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Cascade: a R package to study, predict and simulate the diffusion of a signal through a temporal gene network.

SUMMARY: Temporal gene interactions, in response to environmental stress, form a complex system that can be efficiently described using gene regulatory networks. They allow highlighting the more influential genes and spotting some targets for biological intervention experiments. Despite that many reverse engineering tools have been designed, the Cascade package is an integrated solution adding several new and original key features such as the ability to predict changes in gene expressions after a biological perturbation in the network and graphical outputs that allow monitoring the spread of a signal through the network. AVAILABILITY AND IMPLEMENTATION: The R package Cascade is available online at http://www-math.u-strasbg.fr/genpred/spip.php?rubrique4.

Effect of continuous positive airway pressure treatment on short-term memory performance over 24 h of sustained wakefulness in patients with obstructive sleep apnea-hypopnea syndrome.

BACKGROUND: Few studies have examined the impact of continuous positive airway pressure (CPAP) therapy on short-term memory (STM) over sustained wakefulness in patients with obstructive sleep apnea-hypopnea syndrome (OSAHS). We have investigated if impaired STM can be reversed by CPAP treatment in a 24-h sustained wakefulness paradigm. METHODS: Our follow-up study was conducted with repeated-memory tasks within 12 OSAHS patients and 10 healthy controls who underwent three 32-h sessions, one before CPAP (T0) and the second (T3) and the third (T6), after 3 and 6 months of treatment, respectively, for OSAHS patients. Each session included one night of sleep followed by 24h of sustained wakefulness, during which both groups performed STM tasks including both digit span (DS) and Sternberg tasks. RESULTS: Untreated OSAHS patients had no deficit in the forward DS task measuring immediate memory but were impaired in STM, especially working memory assessed by the complex Sternberg task and the backward DS. However, only performance in the latter was improved after 6 months of CPAP treatment. CONCLUSIONS: Because the high level of memory scanning required high speed in information processing, persistent impairment on the complex Sternberg task may be attributable to working memory slowing, possibly enhanced by sustained wakefulness.

Reverse-engineering the genetic circuitry of a cancer cell with predicted intervention in chronic lymphocytic leukemia.

Cellular behavior is sustained by genetic programs that are progressively disrupted in pathological conditions--notably, cancer. High-throughput gene expression profiling has been used to infer statistical models describing these cellular programs, and development is now needed to guide orientated modulation of these systems. Here we develop a regression-based model to reverse-engineer a temporal genetic program, based on relevant patterns of gene expression after cell stimulation. This method integrates the temporal dimension of biological rewiring of genetic programs and enables the prediction of the effect of targeted gene disruption at the system level. We tested the performance accuracy of this model on synthetic data before reverse-engineering the response of primary cancer cells to a proliferative (protumorigenic) stimulation in a multistate leukemia biological model (i.e., chronic lymphocytic leukemia). To validate the ability of our method to predict the effects of gene modulation on the global program, we performed an intervention experiment on a targeted gene. Comparison of the predicted and observed gene expression changes demonstrates the possibility of predicting the effects of a perturbation in a gene regulatory network, a first step toward an orientated intervention in a cancer cell genetic program.

Behavioural adjustment in response to increased predation risk: a study in three duck species.

Predation directly triggers behavioural decisions designed to increase immediate survival. However, these behavioural modifications can have long term costs. There is therefore a trade-off between antipredator behaviours and other activities. This trade-off is generally considered between vigilance and only one other behaviour, thus neglecting potential compensations. In this study, we considered the effect of an increase in predation risk on the diurnal time-budget of three captive duck species during the wintering period. We artificially increased predation risk by disturbing two groups of 14 mallard and teals at different frequencies, and one group of 14 tufted ducks with a radio-controlled stressor. We recorded foraging, vigilance, preening and sleeping durations the week before, during and after disturbance sessions. Disturbed groups were compared to an undisturbed control group. We showed that in all three species, the increase in predation risk resulted in a decrease in foraging and preening and led to an increase in sleeping. It is worth noting that contrary to common observations, vigilance did not increase. However, ducks are known to be vigilant while sleeping. This complex behavioural adjustment therefore seems to be optimal as it may allow ducks to reduce their predation risk. Our results highlight the fact that it is necessary to encompass the whole individual time-budget when studying behavioural modifications under predation risk. Finally, we propose that studies of behavioural time-budget changes under predation risk should be included in the more general framework of the starvation-predation risk trade-off.