BITES: balanced individual treatment effect for survival data.

MOTIVATION: Estimating the effects of interventions on patient outcome is one of the key aspects of personalized medicine. Their inference is often challenged by the fact that the training data comprises only the outcome for the administered treatment, and not for alternative treatments (the so-called counterfactual outcomes). Several methods were suggested for this scenario based on observational data, i.e. data where the intervention was not applied randomly, for both continuous and binary outcome variables. However, patient outcome is often recorded in terms of time-to-event data, comprising right-censored event times if an event does not occur within the observation period. Albeit their enormous importance, time-to-event data are rarely used for treatment optimization. We suggest an approach named BITES (Balanced Individual Treatment Effect for Survival data), which combines a treatment-specific semi-parametric Cox loss with a treatment-balanced deep neural network; i.e. we regularize differences between treated and non-treated patients using Integral Probability Metrics (IPM). RESULTS: We show in simulation studies that this approach outperforms the state of the art. Furthermore, we demonstrate in an application to a cohort of breast cancer patients that hormone treatment can be optimized based on six routine parameters. We successfully validated this finding in an independent cohort. AVAILABILITY AND IMPLEMENTATION: We provide BITES as an easy-to-use python implementation including scheduled hyper-parameter optimization (https://github.com/sschrod/BITES). The data underlying this article are available in the CRAN repository at https://rdrr.io/cran/survival/man/gbsg.html and https://rdrr.io/cran/survival/man/rotterdam.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

SPARC-positive macrophages are the superior prognostic factor in the microenvironment of diffuse large B-cell lymphoma and independent of MYC rearrangement and double-/triple-hit status.

BACKGROUND: Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous disease with respect to outcome. Features of the tumor microenvironment (TME) are associated with prognosis when assessed by gene expression profiling. However, it is uncertain whether assessment of the microenvironment can add prognostic information to the most relevant and clinically well-established molecular subgroups when analyzed by immunohistochemistry (IHC). PATIENTS AND METHODS: We carried out a histopathologic analysis of biomarkers related to TME in a very large cohort (n = 455) of DLBCL treated in prospective trials and correlated with clinicopathologic and molecular data, including chromosomal rearrangements and gene expression profiles for cell-of-origin and TME. RESULTS: The content of PD1+, FoxP3+ and CD8+, as well as vessel density, was not associated with outcome. However, we found a low content of CD68+ macrophages to be associated with inferior progression-free survival (PFS) and overall survival (OS; P = 0.023 and 0.040, respectively) at both univariable and multivariable analyses, adjusted for the factors of the International Prognostic Index (IPI), MYC break and BCL2/MYC and BCL6/MYC double-hit status. The subgroup of PDL1+ macrophages was not associated with survival. Instead, secreted protein acidic and cysteine rich (SPARC)-positive macrophages were identified as the subtype of macrophages most associated with survival. SPARC-positive macrophages and stromal cells directly correlated with favorable PFS and OS (both, P[log rank] <0.001, P[trend] < 0.001). The association of SPARC with prognosis was independent of the factors of the IPI, MYC double-/triple-hit status, Bcl2/c-myc double expression, cell-of-origin subtype and a recently published gene expression signature [lymphoma-associated macrophage interaction signature (LAMIS)]. CONCLUSIONS: SPARC expression in the TME detected by a single IHC staining with fair-to-good interobserver reproducibility is a powerful prognostic parameter. Thus SPARC expression is a strong candidate for risk assessment in DLBCL in daily practice.

Molecular signatures that can be transferred across different omics platforms.

MOTIVATION: Molecular signatures for treatment recommendations are well researched. Still it is challenging to apply them to data generated by different protocols or technical platforms. RESULTS: We analyzed paired data for the same tumors (Burkitt lymphoma, diffuse large B-cell lymphoma) and features that had been generated by different experimental protocols and analytical platforms including the nanoString nCounter and Affymetrix Gene Chip transcriptomics as well as the SWATH and SRM proteomics platforms. A statistical model that assumes independent sample and feature effects accounted for 69-94% of technical variability. We analyzed how variability is propagated through linear signatures possibly affecting predictions and treatment recommendations. Linear signatures with feature weights adding to zero were substantially more robust than unbalanced signatures. They yielded consistent predictions across data from different platforms, both for transcriptomics and proteomics data. Similarly stable were their predictions across data from fresh frozen and matching formalin-fixed paraffin-embedded human tumor tissue. AVAILABILITY AND IMPLEMENTATION: The R-package 'zeroSum' can be downloaded at https://github.com/rehbergT/zeroSum . Complete data and R codes necessary to reproduce all our results can be received from the authors upon request. CONTACT: rainer.spang@ur.de.

Reference point insensitive molecular data analysis.

MOTIVATION: In biomedicine, every molecular measurement is relative to a reference point, like a fixed aliquot of RNA extracted from a tissue, a defined number of blood cells, or a defined volume of biofluid. Reference points are often chosen for practical reasons. For example, we might want to assess the metabolome of a diseased organ but can only measure metabolites in blood or urine. In this case, the observable data only indirectly reflects the disease state. The statistical implications of these discrepancies in reference points have not yet been discussed. RESULTS: Here, we show that reference point discrepancies compromise the performance of regression models like the LASSO. As an alternative, we suggest zero-sum regression for a reference point insensitive analysis. We show that zero-sum regression is superior to the LASSO in case of a poor choice of reference point both in simulations and in an application that integrates intestinal microbiome analysis with metabolomics. Moreover, we describe a novel coordinate descent based algorithm to fit zero-sum elastic nets. AVAILABILITY AND IMPLEMENTATION: The R-package "zeroSum" can be downloaded at https://github.com/rehbergT/zeroSum Moreover, we provide all R-scripts and data used to produce the results of this manuscript as Supplementary Material CONTACT: Michael.Altenbuchinger@ukr.de, Thorsten.Rehberg@ukr.de and Rainer.Spang@ukr.deSupplementary information: Supplementary material is available at Bioinformatics online.

Gene expression and copy number profiling of follicular lymphoma biopsies from patients treated with first-line rituximab without chemotherapy.

The Nordic Lymphoma Study Group has performed two randomized clinical trials with chemotherapy-free first-line treatment (rituximab +/- interferon) in follicular lymphoma (FL), with 73% of patients alive and 38% without any need of chemotherapy after 10.6 years median follow-up. In order to identify predictive markers, that may also serve as therapeutic targets, gene expression- and copy number profiles were obtained from 97 FL patients using whole genome microarrays. Copy number alterations (CNAs) were identified, e.g. by GISTIC. Cox Lasso Regression and Lasso logistic regression were used to determine molecular features predictive of time to next therapy (TTNT). A few molecular changes were associated with TTNT (e.g. increased expression of INPP5B, gains in 12q23/q24), but were not significant after adjusting for multiple testing. Our findings suggest that there are no strong determinants of patient outcome with respect to GE data and CNAs in FL patients treated with a chemotherapy-free regimen (i.e. rituximab +/- interferon).

Gene expression shift towards normal B cells, decreased proliferative capacity and distinct surface receptors characterize leukemic blasts persisting during induction therapy in childhood acute lymphoblastic leukemia.

In childhood acute lymphoblastic leukemia (ALL), persistence of leukemic blasts during therapy is of crucial prognostic significance. In the present study, we address molecular and cell biologic features of blasts persisting after 1 week of induction glucocorticoid therapy. Genome-wide gene expression analysis of leukemic samples from precursor B-cell ALL patients (n=18) identified a set of genes differentially expressed in blasts at diagnosis day 0 (d0) and persisting on day 8 (d8). Expression changes indicate a shift towards mature B cells, inhibition of cell cycling and increased expression of adhesion (CD11b/ITGAM) and cytokine (CD119/IFNGR1) receptors. A direct comparison with normal B cells, which are largely therapy resistant, confirmed the differentiation shift at the mRNA (n=10) and protein (n=109) levels. Flow cytometric analysis in independent cohorts of patients confirmed both a decreased proliferative activity (n=13) and the upregulation of CD11b and CD119 (n=29) in d8 blasts. The differentiation shift and low proliferative activity in d8 blasts may account for the persistence of blasts during therapy and affect their sensitivity to further therapeutic treatment. CD11b and CD119 are potential specific markers for d8 blast persistence and detection of minimal residual disease, which warrant further investigation.

Role for E2F in control of both DNA replication and mitotic functions as revealed from DNA microarray analysis.

We have used high-density DNA microarrays to provide an analysis of gene regulation during the mammalian cell cycle and the role of E2F in this process. Cell cycle analysis was facilitated by a combined examination of gene control in serum-stimulated fibroblasts and cells synchronized at G(1)/S by hydroxyurea block that were then released to proceed through the cell cycle. The latter approach (G(1)/S synchronization) is critical for rigorously maintaining cell synchrony for unambiguous analysis of gene regulation in later stages of the cell cycle. Analysis of these samples identified seven distinct clusters of genes that exhibit unique patterns of expression. Genes tend to cluster within these groups based on common function and the time during the cell cycle that the activity is required. Placed in this context, the analysis of genes induced by E2F proteins identified genes or expressed sequence tags not previously described as regulated by E2F proteins; surprisingly, many of these encode proteins known to function during mitosis. A comparison of the E2F-induced genes with the patterns of cell growth-regulated gene expression revealed that virtually all of the E2F-induced genes are found in only two of the cell cycle clusters; one group was regulated at G(1)/S, and the second group, which included the mitotic activities, was regulated at G(2). The activation of the G(2) genes suggests a broader role for E2F in the control of both DNA replication and mitotic activities.

Molecular diagnosis. Classification, model selection and performance evaluation.

OBJECTIVES: We discuss supervised classification techniques applied to medical diagnosis based on gene expression profiles. Our focus lies on strategies of adaptive model selection to avoid overfitting in high-dimensional spaces. METHODS: We introduce likelihood-based methods, classification trees, support vector machines and regularized binary regression. For regularization by dimension reduction, we describe feature selection methods: feature filtering, feature shrinkage and wrapper approaches. In small sample-size situations efficient methods of data re-use are needed to assess the predictive power of a model. We discuss two issues in using cross-validation: the difference between in-loop and out-of-loop feature selection, and estimating model parameters in nested-loop cross-validation. RESULTS: Gene selection does not reduce the dimensionality of the model. Tuning parameters enable adaptive model selection. The feature selection bias is a common pitfall in performance evaluation. Model selection and performance evaluation can be combined by nested-loop cross-validation. CONCLUSIONS: Classification of microarrays is prone to overfitting. A rigorous and unbiased assessment of the predictive power of the model is a must.

Gene expression profiling suggests primary central nervous system lymphomas to be derived from a late germinal center B cell.

To characterize the molecular origin of primary lymphomas of the central nervous system (PCNSL), 21 PCNSLs of immunocompetent patients were investigated by microarray-based gene expression profiling. Comparison of the transcriptional profile of PCNSL with various normal and neoplastic B-cell subsets demonstrated PCNSL (i) to display gene expression patterns most closely related to late germinal center B cells, (ii) to display a gene expression profile similar to systemic diffuse large B-cell lymphomas (DLBCLs) and (iii) to be in part assigned to the activated B-cell-like (ABC) or the germinal center B-cell-like (GCB) subtype of DLBCL.

Pathway activation patterns in diffuse large B-cell lymphomas.

Deregulation of cell signaling pathways controlling cell growth and cell survival is a common feature of all cancers. Although a core repertoire of oncogenic mechanisms is widely conserved between various malignancies, the constellation of pathway activities can vary even in patients with the same malignant disease. Modern molecularly targeted cancer drugs intervene in cell signaling compensating for pathway deregulation. Hence characterizing tumors with respect to pathway activation will become crucial for treatment decisions. Here we have used semi-supervised machine learning methodology to generate signatures of eight oncogene-inducible pathways, which are conserved across epithelial and lymphoid tissues. We combined them to patterns of pathway activity called PAPs for pathway activation patterns and searched for them in 220 morphologically, immunohistochemically and genetically well-characterized mature aggressive B-cell lymphomas including 134 cases with clinical data available. Besides Burkitt lymphoma, which was characterized by a unique pattern, the PAPs identified four distinct groups of mature aggressive B-cell lymphomas across independent gene expression studies with distinct biological characteristics, genetic aberrations and prognosis. We confirmed our findings through cross-platform analysis in an independent data set of 303 mature aggressive B-cell lymphomas.

Statistics of large-scale sequence searching.

MOTIVATION: Database search programs such as FASTA, BLAST or a rigorous Smith-Waterman algorithm produce lists of database entries, which are assumed to be related to the query. The computation of statistical significance of similarity scores is well established for single pairs of sequences and using purely random models. However, the multi-trial context of a database search poses new problems. The credibility of a certain score obtained in a database search decreases with the amount of data that is compared. To improve p-value computation for database search experiments, statistical properties of the databases, such as the distribution of sequence length and effects induced by frequently repeated sequence patterns, need to be taken into account. RESULTS: We investigated the SWISS-PROT protein database Release 31.0 running extensive simulations of database searches. A discrepancy is observed between the theoretical predictions and the empirical distribution. To correct for this, we evaluate the statistical significance of scores in the context of a database search by a contrasting semi-random model. This model enhances purely random models by one additional parameter reflecting individual statistical properties of real databases. We call this parameter the effective size of the database. CONTACT: r.spang@dkfz-heidelberg.de;m.vingron@dkfz-hei del berg.de

Limits of homology detection by pairwise sequence comparison.

MOTIVATION: Noise in database searches resulting from random sequence similarities increases as the databases expand rapidly. The noise problems are not a technical shortcoming of the database search programs, but a logical consequence of the idea of homology searches. The effect can be observed in simulation experiments. RESULTS: We have investigated noise levels in pairwise alignment based database searches. The noise levels of 38 releases of the SwissProt database, display perfect logarithmic growth with the total length of the databases. Clustering of real biological sequences reduces noise levels, but the effect is marginal.

Sequence database search using jumping alignments.

We describe a new algorithm for amino acid sequence classification and the detection of remote homologues. The rationale is to exploit both vertical and horizontal information of a multiple alignment in a well balanced manner. This is in contrast to established methods like profiles and hidden Markov models which focus on vertical information as they model the columns of the alignment independently. In our setting, we want to select from a given database of "candidate sequences" those proteins that belong to a given superfamily. In order to do so, each candidate sequence is separately tested against a multiple alignment of the known members of the superfamily by means of a new jumping alignment algorithm. This algorithm is an extension of the Smith-Waterman algorithm and computes a local alignment of a single sequence and a multiple alignment. In contrast to traditional methods, however, this alignment is not based on a summary of the individual columns of the multiple alignment. Rather, the candidate sequence at each position is aligned to one sequence of the multiple alignment, called the "reference sequence". In addition, the reference sequence may change within the alignment, while each such jump is penalized. To evaluate the discriminative quality of the jumping alignment algorithm, we compared it to hidden Markov models on a subset of the SCOP database of protein domains. The discriminative quality was assessed by counting the number of false positives that ranked higher than the first true positive (FP-count). For moderate FP-counts above five, the number of successful searches with our method was considerably higher than with hidden Markov models.

Malformations in 10,000 consecutive births in Tunis.

Malformations were assessed in 10,000 consecutively born infants, dead or alive, at the Wassila Bourgiba Maternity Hospital in Tunis. The medical and social history including the rate of consanguinity was studied in the malformed group as well as in a control group of 229 infants. Three hundred and ninety-six infants were malformed; 248 had major malformations and 148 had minor ones. Thirteen per cent of the stillborn were malformed compared to 3.7% of the liveborn. The rates of most specific malformations were comparable to those in other studies but a relatively high rate of neural tube defects, 2.2/1000, can be noted. There is a significant overrepresentation of consanguinity (65%) in parents of non syndromic multi-malformed infants.

Predicting the clinical status of human breast cancer by using gene expression profiles.

Prognostic and predictive factors are indispensable tools in the treatment of patients with neoplastic disease. For the most part, such factors rely on a few specific cell surface, histological, or gross pathologic features. Gene expression assays have the potential to supplement what were previously a few distinct features with many thousands of features. We have developed Bayesian regression models that provide predictive capability based on gene expression data derived from DNA microarray analysis of a series of primary breast cancer samples. These patterns have the capacity to discriminate breast tumors on the basis of estrogen receptor status and also on the categorized lymph node status. Importantly, we assess the utility and validity of such models in predicting the status of tumors in crossvalidation determinations. The practical value of such approaches relies on the ability not only to assess relative probabilities of clinical outcomes for future samples but also to provide an honest assessment of the uncertainties associated with such predictive classifications on the basis of the selection of gene subsets for each validation analysis. This latter point is of critical importance in the ability to apply these methodologies to clinical assessment of tumor phenotype.