Machine Learning Identifies Stemness Features Associated with Oncogenic Dedifferentiation.

Cancer progression involves the gradual loss of a differentiated phenotype and acquisition of progenitor and stem-cell-like features. Here, we provide novel stemness indices for assessing the degree of oncogenic dedifferentiation. We used an innovative one-class logistic regression (OCLR) machine-learning algorithm to extract transcriptomic and epigenetic feature sets derived from non-transformed pluripotent stem cells and their differentiated progeny. Using OCLR, we were able to identify previously undiscovered biological mechanisms associated with the dedifferentiated oncogenic state. Analyses of the tumor microenvironment revealed unanticipated correlation of cancer stemness with immune checkpoint expression and infiltrating immune cells. We found that the dedifferentiated oncogenic phenotype was generally most prominent in metastatic tumors. Application of our stemness indices to single-cell data revealed patterns of intra-tumor molecular heterogeneity. Finally, the indices allowed for the identification of novel targets and possible targeted therapies aimed at tumor differentiation.

Unveiling the Molecular Mechanism of Trastuzumab Resistance in SKBR3 and BT474 Cell Lines for HER2 Positive Breast Cancer.

HER2-positive breast cancer is one of the most prevalent forms of cancer among women worldwide. Generally, the molecular characteristics of this breast cancer include activation of human epidermal growth factor receptor-2 (HER2) and hormone receptor activation. HER2-positive is associated with a higher death rate, which led to the development of a monoclonal antibody called trastuzumab, specifically targeting HER2. The success rate of HER2-positive breast cancer treatment has been increased; however, drug resistance remains a challenge. This fact motivated us to explore the underlying molecular mechanisms of trastuzumab resistance. For this purpose, a two-fold approach was taken by considering well-known breast cancer cell lines SKBR3 and BT474. In the first fold, trastuzumab treatment doses were optimized separately for both cell lines. This was done based on the proliferation rate of cells in response to a wide variety of medication dosages. Thereafter, each cell line was cultivated with a steady dosage of herceptin for several months. During this period, six time points were selected for further in vitro analysis, ranging from the untreated cell line at the beginning to a fully resistant cell line at the end of the experiment. In the second fold, nucleic acids were extracted for further high throughput-based microarray experiments of gene and microRNA expression. Such expression data were further analyzed in order to infer the molecular mechanisms involved in the underlying development of trastuzumab resistance. In the list of differentially expressed genes and miRNAs, multiple genes (e.g., BIRC5, E2F1, TFRC, and USP1) and miRNAs (e.g., hsa miR 574 3p, hsa miR 4530, and hsa miR 197 3p) responsible for trastuzumab resistance were found. Downstream analysis showed that TFRC, E2F1, and USP1 were also targeted by hsa-miR-8485. Moreover, it indicated that miR-4701-5p was highly expressed as compared to TFRC in the SKBR3 cell line. These results unveil key genes and miRNAs as molecular regulators for trastuzumab resistance.

Using an early outcome as the sole source of information of interim decisions regarding treatment effect on a long-term endpoint: The non-Gaussian case.

In randomized clinical trials that use a long-term efficacy endpoint, the follow-up time necessary to observe the endpoint may be substantial. In such trials, an attractive option is to consider an interim analysis based solely on an early outcome that could be used to expedite the evaluation of treatment's efficacy. Garcia Barrado et al. (Pharm Stat. 2022; 21: 209-219) developed a methodology that allows introducing such an early interim analysis for the case when both the early outcome and the long-term endpoint are normally-distributed, continuous variables. We extend the methodology to any combination of the early-outcome and long-term-endpoint types. As an example, we consider the case of a binary outcome and a time-to-event endpoint. We further evaluate the potential gain in operating characteristics (power, expected trial duration, and expected sample size) of a trial with such an interim analysis in function of the properties of the early outcome as a surrogate for the long-term endpoint.

Minimization in randomized clinical trials.

In randomized trials, comparability of the treatment groups is ensured through allocation of treatments using a mechanism that involves some random element, thus controlling for confounding of the treatment effect. Completely random allocation ensures comparability between the treatment groups for all known and unknown prognostic factors. For a specific trial, however, imbalances in prognostic factors among the treatment groups may occur. Although accidental bias can be avoided in the presence of such imbalances by stratifying the analysis, most trialists, regulatory agencies, and other stakeholders prefer a balanced distribution of prognostic factors across the treatment groups. Some procedures attempt to achieve balance in baseline covariates, by stratifying the allocation for these covariates, or by dynamically adapting the allocation using covariate information during the trial (covariate-adaptive procedures). In this Tutorial, the performance of minimization, a popular covariate-adaptive procedure, is compared with two other commonly used procedures, completely random allocation and stratified blocked designs. Using individual patient data of 2 clinical trials (in advanced ovarian cancer and age-related macular degeneration), the procedures are compared in terms of operating characteristics (using asymptotic and randomization tests), predictability of treatment allocation, and achieved balance. Fifty actual trials of various sizes that applied minimization for treatment allocation are used to investigate the achieved balance. Implementation issues of minimization are described. Minimization procedures are useful in all trials but especially when (1) many major prognostic factors are known, (2) many centers of different sizes accrue patients, or (3) the trial sample size is moderate.

Machine learning approach for the prediction of the number of sulphur atoms in peptides using the theoretical aggregated isotope distribution.

RATIONALE: The observed isotope distribution is an important attribute for the identification of peptides and proteins in mass spectrometry-based proteomics. Sulphur atoms have a very distinctive elemental isotope definition, and therefore, the presence of sulphur atoms has a substantial effect on the isotope distribution of biomolecules. Hence, knowledge of the number of sulphur atoms can improve the identification of peptides and proteins. METHODS: In this paper, we conducted a theoretical investigation on the isotope properties of sulphur-containing peptides. We proposed a gradient boosting approach to predict the number of sulphur atoms based on the aggregated isotope distribution. We compared prediction accuracy and assessed the predictive power of the features using the mass and isotope abundance information from the first three, five and eight aggregated isotope peaks. RESULTS: Mass features alone are not sufficient to accurately predict the number of sulphur atoms. However, we reach near-perfect prediction when we include isotope abundance features. The abundance ratios of the eighth and the seventh, the fifth and the fourth, and the third and the second aggregated isotope peaks are the most important abundance features. The mass difference between the eighth, the fifth or the third aggregated isotope peaks and the monoisotopic peak are the most predictive mass features. CONCLUSIONS: Based on the validation analysis it can be concluded that the prediction of the number of sulphur atoms based on the isotope profile fails, because the isotope ratios are not measured accurately. These results indicate that it is valuable for future instrument developments to focus more on improving spectral accuracy to measure peak intensities of higher-order isotope peaks more accurately.

Over-accrual in Bayesian adaptive trials with continuous futility stopping.

BACKGROUND: We explore frequentist operating characteristics of a Bayesian adaptive design that allows continuous early stopping for futility. In particular, we focus on the power versus sample size relationship when more patients are accrued than originally planned. METHODS: We consider the case of a phase II single-arm study and a Bayesian phase II outcome-adaptive randomization design. For the former, analytical calculations are possible; for the latter, simulations are conducted. RESULTS: Results for both cases show a decrease in power with an increasing sample size. It appears that this effect is due to the increasing cumulative probability of incorrectly stopping for futility. CONCLUSION: The increase in cumulative probability of incorrectly stopping for futility is related to the continuous nature of the early stopping, which increases the number of interim analyses with accrual. The issue can be addressed by, for instance, delaying the start of testing for futility, reducing the number of futility tests to be performed or by setting stricter criteria for concluding futility.

Generalized pairwise comparisons for censored data: An overview.

The method of generalized pairwise comparisons (GPC) is an extension of the well-known nonparametric Wilcoxon-Mann-Whitney test for comparing two groups of observations. Multiple generalizations of Wilcoxon-Mann-Whitney test and other GPC methods have been proposed over the years to handle censored data. These methods apply different approaches to handling loss of information due to censoring: ignoring noninformative pairwise comparisons due to censoring (Gehan, Harrell, and Buyse); imputation using estimates of the survival distribution (Efron, Péron, and Latta); or inverse probability of censoring weighting (IPCW, Datta and Dong). Based on the GPC statistic, a measure of treatment effect, the "net benefit," can be defined. It quantifies the difference between the probabilities that a randomly selected individual from one group is doing better than an individual from the other group. This paper aims at evaluating GPC methods for censored data, both in the context of hypothesis testing and estimation, and providing recommendations related to their choice in various situations. The methods that ignore uninformative pairs have comparable power to more complex and computationally demanding methods in situations of low censoring, and are slightly superior for high proportions (>40%) of censoring. If one is interested in estimation of the net benefit, Harrell's c index is an unbiased estimator if the proportional hazards assumption holds. Otherwise, the imputation (Efron or Peron) or IPCW (Datta, Dong) methods provide unbiased estimators in case of proportions of drop-out censoring up to 60%.

Surrogacy Beyond Prognosis: The Importance of "Trial-Level" Surrogacy.

Many candidate surrogate endpoints are currently assessed using a 2-level statistical approach, which consists in checking whether (1) the potential surrogate is associated with the final endpoint in individual patients and (2) the effect of treatment on the surrogate can be used to reliably predict the effect of treatment on the final endpoint. In some situations, condition (1) is fulfilled but condition (2) is not. We use concepts of causal inference to explain this apparently paradoxical situation, illustrating this review with 2 contrasting examples in operable breast cancer: the example of pathological complete response (pCR) and that of disease-free survival (DFS). In a previous meta-analysis, pCR has been shown to be a strong and independent prognostic factor for event-free survival (EFS) and overall survival (OS) after neoadjuvant treatment of operable breast cancer. Yet, in randomized trials, the effects of experimental treatments on pCR have not translated into predictable effects on EFS or OS, making pCR an "individual-level" surrogate, but not a "trial-level" surrogate. In contrast, DFS has been shown to be an acceptable surrogate for OS at both the individual and trial levels in early, HER2-positive breast cancer. The distinction between the prognostic and predictive roles of a tentative surrogate, not always made in the literature, avoids unnecessary confusion and allows better understanding of what it takes to validate a surrogate endpoint that is truly able to replace a final endpoint.

The case against censoring of progression-free survival in cancer clinical trials - A pandemic shutdown as an illustration.

BACKGROUND: Missing data may lead to loss of statistical power and introduce bias in clinical trials. The Covid-19 pandemic has had a profound impact on patient health care and on the conduct of cancer clinical trials. Although several endpoints may be affected, progression-free survival (PFS) is of major concern, given its frequent use as primary endpoint in advanced cancer and the fact that missed radiographic assessments are to be expected. The recent introduction of the estimand framework creates an opportunity to define more precisely the target of estimation and ensure alignment between the scientific question and the statistical analysis. METHODS: We used simulations to investigate the impact of two basic approaches for handling missing tumor scans due to the pandemic: a "treatment policy" strategy, which consisted in ascribing events to the time they are observed, and a "hypothetical" approach of censoring patients with events during the shutdown period at the last assessment prior to that period. We computed the power of the logrank test, estimated hazard ratios (HR) using Cox models, and estimated median PFS times without and with a hypothetical 6-month shutdown period with no patient enrollment or tumor scans being performed, varying the shutdown starting times. RESULTS: Compared with the results in the absence of shutdown, the "treatment policy" strategy slightly overestimated median PFS proportionally to the timing of the shutdown period, but power was not affected. Except for one specific scenario, there was no impact on the estimated HR. In general, the pandemic had a greater impact on the analyses using the "hypothetical" strategy, which led to decreased power and overestimated median PFS times to a greater extent than the "treatment policy" strategy. CONCLUSION: As a rule, we suggest that the treatment policy approach, which conforms with the intent-to-treat principle, should be the primary analysis to avoid unnecessary loss of power and minimize bias in median PFS estimates.

Semi-parametric accelerated failure-time model: A useful alternative to the proportional-hazards model in cancer clinical trials.

The accelerated failure-time (AFT) model has been long recognized as a useful alternative to the proportional-hazards (PH) model. Semi-parametric AFT model has been known since 1981. Its use has been hampered by the difficulty in solving the estimating equations for the model's coefficients. In recent years, however, important developments have taken place regarding the methods of solving the equations. In this article, we briefly review the developments, focusing mainly on rank-based estimation. We conduct a simulation study that directly focuses on the applicability of the model in the context of (cancer) clinical trials. We also investigate the robustness of the AFT model to the omission of covariates. Finally, we conduct a meta-analysis of multiple clinical trials in gastric cancer to illustrate the benefits of the use of the model in practice.

Operational characteristics of generalized pairwise comparisons for hierarchically ordered endpoints.

The method of generalized pairwise comparisons (GPC) is a multivariate extension of the well-known non-parametric Wilcoxon-Mann-Whitney test. It allows comparing two groups of observations based on multiple hierarchically ordered endpoints, regardless of the number or type of the latter. The summary measure, "net benefit," quantifies the difference between the probabilities that a random observation from one group is doing better than an observation from the opposite group. The method takes into account the correlations between the endpoints. We have performed a simulation study for the case of two hierarchical endpoints to evaluate the impact of their correlation on the type-I error probability and power of the test based on GPC. The simulations show that the power of the GPC test for the primary endpoint is modified if the secondary endpoint is included in the hierarchical GPC analysis. The change in power depends on the correlation between the endpoints. Interestingly, a decrease in power can occur, regardless of whether there is any marginal treatment effect on the secondary endpoint. It appears that the overall power of the hierarchical GPC procedure depends, in a complex manner, on the entire variance-covariance structure of the set of outcomes. Any additional factors (such as thresholds of clinical relevance, drop out, or censoring scheme) will also affect the power and will have to be taken into account when designing a trial based on the hierarchical GPC procedure.

Using an interim analysis based exclusively on an early outcome in a randomized clinical trial with a long-term clinical endpoint.

In RCTs with an interest in a long-term efficacy endpoint, the follow-up time necessary to observe the endpoint may be substantial. In order to reduce the expected duration of such trials, early-outcome data may be collected to enrich an interim analysis aimed at stopping the trial early for efficacy. We propose to extend such a design with an additional interim analysis using solely early-outcome data in order to expedite the evaluation of treatment's efficacy. We evaluate the potential gain in operating characteristics (power, expected trial duration, and expected sample size) when introducing such an early interim analysis, in function of the properties of the early outcome as a surrogate for the long-term endpoint. In the context of a longitudinal age-related macular degeneration (ARMD) ophthalmology trial, results show potentially substantial gains in both the expected trial duration and the expected sample size. A prerequisite, though, is that the treatment effect on the early outcome has to be strongly correlated with the treatment effect on the long-term endpoint, that is, that the early outcome is a validated surrogate for the long-term endpoint.

Predicting the number of sulfur atoms in peptides and small proteins based on the observed aggregated isotope distribution.

RATIONALE: Identification of peptides and proteins is a challenging task in mass spectrometry-based proteomics. Knowledge of the number of sulfur atoms can improve the identification of peptides and proteins. METHODS: In this article, we propose a method for the prediction of S-atoms based on the aggregated isotope distribution. The Mahalanobis distance is used as dissimilarity measure to compare mass- and intensity-based features from the observed and theoretical isotope distributions. RESULTS: The relative abundance of the second and the third aggregated isotopic variants (as compared to the monoisotopic one) and the mass difference between the second and third aggregated isotopic variants are the most important features to predict the number of S-atoms. CONCLUSIONS: The mass and intensity accuracies of the observed aggregated isotopic variants are insufficient to accurately predict the number of atoms. However, using a limited set of predictions for a peptide, rather than predicting a single number of S-atoms, has a reasonably high prediction accuracy.

Bayesian biomarker-driven outcome-adaptive randomization with an imperfect biomarker assay.

OBJECTIVE: We investigate the impact of biomarker assay's accuracy on the operating characteristics of a Bayesian biomarker-driven outcome-adaptive randomization design. METHODS: In a simulation study, we assume a trial with two treatments, two biomarker-based strata, and a binary clinical outcome (response). Pbt denotes the probability of response for treatment t (t = 0 or 1) in biomarker stratum (b = 0 or 1). Four different scenarios in terms of true underlying response probabilities are considered: a null (P00 = P01 = 0.25, P10 = P11= 0.25) and consistent (P00 = P10 = 0.25, P01 = 0.5) treatment effect scenario, as well as a quantitative (P00 = P01 = P10 = 0.25, P11 = 0.5) and a qualitative (P00 = P11 = 0.5, P01 = P10 = 0.25) stratum-treatment interaction. For each scenario, we compare the case of a perfect with the case of an imperfect biomarker assay with sensitivity and specificity of 0.8 and 0.7, respectively. In addition, biomarker-positive prevalence values P(B = 1) = 0.2 and 0.5 are investigated. RESULTS: Results show that the use of an imperfect assay affects the operational characteristics of the Bayesian biomarker-based outcome-adaptive randomization design. In particular, the misclassification causes a substantial reduction in power accompanied by a considerable increase in the type-I error probability. The magnitude of these effects depends on the sensitivity and specificity of the assay, as well as on the distribution of the biomarker in the patient population. CONCLUSION: With an imperfect biomarker assay, the decision to apply a biomarker-based outcome-adaptive randomization design may require careful reflection.

Status of the alveolar bone after autotransplantation of developing premolars to the anterior maxilla assessed by CBCT measurements.

BACKGROUND/AIMS: Autotransplantation of developing premolars is an established treatment to replace missing teeth in the anterior maxilla in growing patients with a reported success rate of over 90%. The normal shape of the alveolus is observed after transplantation, but data on the presence and amount of alveolar bone after healing has not been previously reported. The aim of this study was to look for potential differences in alveolar bone dimensions between sites where autotransplanted premolars replaced missing incisors and control sites of contralateral incisors. MATERIAL/METHODS: There were 11 patients aged between 10 and 12 years five months (mean age: 10 years and 7 months) who underwent autotransplantation of a premolar to replace a central incisor. Cone Beam Computed Tomography (CBCT) performed at least 1 year after transplantation served to evaluate bone at sites of autotransplanted premolars and controls (contralateral maxillary central incisor). The thickness of the labial bone, plus the height and width of the alveolar process were measured on scans and compared at transplant and control sites. RESULTS: Mean thicknesses of the labial bone at the transplant and control sites were 0.78 mm and 0.82 mm respectively. Mean alveolar bone height was 15.15 mm at the transplant sites and 15.12 mm at the control sites. The mean marginal thickness of the alveolus was 7.75 mm at the transplant sites and 7.98 mm at the control sites. Mean thicknesses of the alveolus for half of its vertical dimension at the transplant and control sites were 7.54 mm and 8.03 mm, respectively. CONCLUSION: The mean values of bone thickness, width and height of the alveolar process at sites of transplanted premolars were comparable to the mean values for the control incisors. Successful autotransplantation of developing premolars to replace missing central incisors allowed preservation of alveolar bone in the anterior maxilla.

A multivariate linear model for investigating the association between gene-module co-expression and a continuous covariate.

A way to enhance our understanding of the development and progression of complex diseases is to investigate the influence of cellular environments on gene co-expression (i.e. gene-pair correlations). Often, changes in gene co-expression are investigated across two or more biological conditions defined by categorizing a continuous covariate. However, the selection of arbitrary cut-off points may have an influence on the results of an analysis. To address this issue, we use a general linear model (GLM) for correlated data to study the relationship between gene-module co-expression and a covariate like metabolite concentration. The GLM specifies the gene-pair correlations as a function of the continuous covariate. The use of the GLM allows for investigating different (linear and non-linear) patterns of co-expression. Furthermore, the modeling approach offers a formal framework for testing hypotheses about possible patterns of co-expression. In our paper, a simulation study is used to assess the performance of the GLM. The performance is compared with that of a previously proposed GLM that utilizes categorized covariates. The versatility of the model is illustrated by using a real-life example. We discuss the theoretical issues related to the construction of the test statistics and the computational challenges related to fitting of the proposed model.

Disease-free survival as a surrogate for overall survival in patients with HER2-positive, early breast cancer in trials of adjuvant trastuzumab for up to 1 year: a systematic review and meta-analysis.

BACKGROUND: Although frequently used as a primary endpoint, disease-free survival has not been validated as a surrogate for overall survival in early breast cancer. We investigated this surrogacy in the adjuvant setting of treatment with anti-HER2 antibodies. METHODS: In a systematic review and meta-analysis, we identified published and non-published randomised controlled trials with completed accrual and available disease-free survival and overall survival results for the intention-to-treat population as of September 2016. Bibliographic databases (MEDLINE, Embase, and Cochrane Central Register of Controlled Trials), clinical trial registries (Clinicaltrials.gov, EU Clinical Trials Register, WHO International Clinical Trials Registry Platform, and PharmNet.Bund), and trial registries from relevant pharmaceutical companies were searched. Eligibility for treatment of HER2-positive early breast cancer required at least one group to have an anti-HER antibody treatment (ie, trastuzumab, pertuzumab, or trastuzumab emtansine) planned for 12 months, and at least one control arm with chemotherapy without the antibody, a lower total dose or duration of the antibody, or observation alone. Units of analysis were contrasts: two-group trials gave rise to one contrast, whereas trials with more than two groups gave rise to more than one contrast. We excluded trials enrolling patients with recurrent, metastatic, or non-invasive disease, and those testing neoadjuvant therapy exclusively. Our primary objective was to estimate patient-level and trial-level correlations between disease-free survival and overall survival. We measured the association between disease-free survival and overall survival using Spearman's correlation coefficient (rs), and the association between hazard ratios (HRs) for disease-free survival and overall survival using R2. We computed the surrogate threshold effect, the maximum HR for disease-free survival that statistically predicts an HR for overall survival less than 1·00 in a future trial. FINDINGS: Eight trials (n=21 480 patients) gave rise to a full set (12 contrasts). Patient-level associations between disease-free and overall survival were strong (rs=0·90 [95% CI 0·89-0·90]). Trial-level associations gave rise to values of R2 of 0·75 (95% CI 0·50-1·00) for the full set. Subgroups defined by nodal status and hormone receptor status yielded qualitatively similar results. Depending on the expected number of deaths in a future trial, the surrogate threshold effects ranged from 0·56 to 0·81, based on the full set. INTERPRETATION: These findings suggest that it is appropriate to continue to use disease-free survival as a surrogate for overall survival in trials in HER-2-positive, early breast cancer. The key limitation of this study is the dependence of its results on the trials included and on the existence of an outlying trial. FUNDING: Roche Pharma AG.

Detection of atypical data in multicenter clinical trials using unsupervised statistical monitoring.

BACKGROUND/AIMS: A risk-based approach to clinical research may include a central statistical assessment of data quality. We investigated the operating characteristics of unsupervised statistical monitoring aimed at detecting atypical data in multicenter experiments. The approach is premised on the assumption that, save for random fluctuations and natural variations, data coming from all centers should be comparable and statistically consistent. Unsupervised statistical monitoring consists of performing as many statistical tests as possible on all trial data, in order to detect centers whose data are inconsistent with data from other centers. METHODS: We conducted simulations using data from a large multicenter trial conducted in Japan for patients with advanced gastric cancer. The actual trial data were contaminated in computer simulations for varying percentages of centers, percentages of patients modified within each center and numbers and types of modified variables. The unsupervised statistical monitoring software was run by a blinded team on the contaminated data sets, with the purpose of detecting the centers with contaminated data. The operating characteristics (sensitivity, specificity and Youden's J-index) were calculated for three detection methods: one using the p-values of individual statistical tests after adjustment for multiplicity, one using a summary of all p-values for a given center, called the Data Inconsistency Score, and one using both of these methods. RESULTS: The operating characteristics of the three methods were satisfactory in situations of data contamination likely to occur in practice, specifically when a single or a few centers were contaminated. As expected, the sensitivity increased for increasing proportions of patients and increasing numbers of variables contaminated. The three methods showed a specificity better than 93% in all scenarios of contamination. The method based on the Data Inconsistency Score and individual p-values adjusted for multiplicity generally had slightly higher sensitivity at the expense of a slightly lower specificity. CONCLUSIONS: The use of brute force (a computer-intensive approach that generates large numbers of statistical tests) is an effective way to check data quality in multicenter clinical trials. It can provide a cost-effective complement to other data-management and monitoring techniques.