Determination of permissive and restraining cancer-associated fibroblast (DeCAF) subtypes.

Cancer-associated fibroblast (CAF) subpopulations in pancreatic ductal adenocarcinoma (PDAC) have been identified using single-cell RNA sequencing (scRNAseq) with divergent characteristics, but their clinical relevance remains unclear. We translate scRNAseq-derived CAF cell-subpopulation-specific marker genes to bulk RNAseq data, and develop a single- sample classifier, DeCAF, for the classification of clinically rest raining and perm issive CAF subtypes. We validate DeCAF in 19 independent bulk transcriptomic datasets across four tumor types (PDAC, mesothelioma, bladder and renal cell carcinoma). DeCAF subtypes have distinct histology features, immune landscapes, and are prognostic and predict response to therapy across cancer types. We demonstrate that DeCAF is clinically replicable and robust for the classification of CAF subtypes in patients for multiple tumor types, providing a better framework for the future development and translation of therapies against permissive CAF subtypes and preservation of restraining CAF subtypes. Significance: We introduce a replicable and robust classifier, DeCAF, that delineates the significance of the role of permissive and restraining CAF subtypes in cancer patients. DeCAF is clinically tractable, prognostic and predictive of treatment response in multiple cancer types and lays the translational groundwork for the preclinical and clinical development of CAF subtype specific therapies.

Tumor Intrinsic Subtypes and Gene Expression Signatures in Early-Stage ERBB2/HER2-Positive Breast Cancer: A Pooled Analysis of CALGB 40601, NeoALTTO, and NSABP B-41 Trials.

Importance: Biologic features may affect pathologic complete response (pCR) and event-free survival (EFS) after neoadjuvant chemotherapy plus ERBB2/HER2 blockade in ERBB2/HER2-positive early breast cancer (EBC). Objective: To define the quantitative association between pCR and EFS by intrinsic subtype and by other gene expression signatures in a pooled analysis of 3 phase 3 trials: CALGB 40601, NeoALTTO, and NSABP B-41. Design, Setting, and Participants: In this retrospective pooled analysis, 1289 patients with EBC received chemotherapy plus either trastuzumab, lapatinib, or the combination, with a combined median follow-up of 5.5 years. Gene expression profiling by RNA sequencing was obtained from 758 samples, and intrinsic subtypes and 618 gene expression signatures were calculated. Data analyses were performed from June 1, 2020, to January 1, 2023. Main Outcomes and Measures: The association of clinical variables and gene expression biomarkers with pCR and EFS were studied by logistic regression and Cox analyses. Results: In the pooled analysis, of 758 women, median age was 49 years, 12% were Asian, 6% Black, and 75% were White. Overall, pCR results were associated with EFS in the ERBB2-enriched (hazard ratio [HR], 0.45; 95% CI, 0.29-0.70; P < .001) and basal-like (HR, 0.19; 95% CI, 0.04-0.86; P = .03) subtypes but not in luminal A or B tumors. Dual trastuzumab plus lapatinib blockade over trastuzumab alone had a trend toward EFS benefit in the intention-to-treat population; however, in the ERBB2-enriched subtype there was a significant and independent EFS benefit of trastuzumab plus lapatinib vs trastuzumab alone (HR, 0.47; 95% CI, 0.27-0.83; P = .009). Overall, 275 of 618 gene expression signatures (44.5%) were significantly associated with pCR and 9 of 618 (1.5%) with EFS. The ERBB2/HER2 amplicon and multiple immune signatures were significantly associated with pCR. Luminal-related signatures were associated with lower pCR rates but better EFS, especially among patients with residual disease and independent of hormone receptor status. There was significant adjusted HR for pCR ranging from 0.45 to 0.81 (higher pCR) and 1.21-1.94 (lower pCR rate); significant adjusted HR for EFS ranged from 0.71 to 0.94. Conclusions and relevance: In patients with ERBB2/HER2-positive EBC, the association between pCR and EFS differed by tumor intrinsic subtype, and the benefit of dual ERBB2/HER2 blockade was limited to ERBB2-enriched tumors. Immune-activated signatures were concordantly associated with higher pCR rates and better EFS, whereas luminal signatures were associated with lower pCR rates.

Efficient computation of high-dimensional penalized generalized linear mixed models by latent factor modeling of the random effects.

Modern biomedical datasets are increasingly high-dimensional and exhibit complex correlation structures. Generalized linear mixed models (GLMMs) have long been employed to account for such dependencies. However, proper specification of the fixed and random effects in GLMMs is increasingly difficult in high dimensions, and computational complexity grows with increasing dimension of the random effects. We present a novel reformulation of the GLMM using a factor model decomposition of the random effects, enabling scalable computation of GLMMs in high dimensions by reducing the latent space from a large number of random effects to a smaller set of latent factors. We also extend our prior work to estimate model parameters using a modified Monte Carlo Expectation Conditional Minimization algorithm, allowing us to perform variable selection on both the fixed and random effects simultaneously. We show through simulation that through this factor model decomposition, our method can fit high-dimensional penalized GLMMs faster than comparable methods and more easily scale to larger dimensions not previously seen in existing approaches.

Differential transcript usage analysis incorporating quantification uncertainty via compositional measurement error regression modeling.

Differential transcript usage (DTU) occurs when the relative expression of multiple transcripts arising from the same gene changes between different conditions. Existing approaches to detect DTU often rely on computational procedures that can have speed and scalability issues as the number of samples increases. Here we propose a new method, CompDTU, that uses compositional regression to model the relative abundance proportions of each transcript that are of interest in DTU analyses. This procedure leverages fast matrix-based computations that make it ideally suited for DTU analysis with larger sample sizes. This method also allows for the testing of and adjustment for multiple categorical or continuous covariates. Additionally, many existing approaches for DTU ignore quantification uncertainty in the expression estimates for each transcript in RNA-seq data. We extend our CompDTU method to incorporate quantification uncertainty leveraging common output from RNA-seq expression quantification tool in a novel method CompDTUme. Through several power analyses, we show that CompDTU has excellent sensitivity and reduces false positive results relative to existing methods. Additionally, CompDTUme results in further improvements in performance over CompDTU with sufficient sample size for genes with high levels of quantification uncertainty, while also maintaining favorable speed and scalability. We motivate our methods using data from the Cancer Genome Atlas Breast Invasive Carcinoma data set, specifically using RNA-seq data from primary tumors for 740 patients with breast cancer. We show greatly reduced computation time from our new methods as well as the ability to detect several novel genes with significant DTU across different breast cancer subtypes.

DNA Mutational Profiling in Patients With Colorectal Cancer Treated With Standard of Care Reveals Differences in Outcome and Racial Distribution of Mutations.

PURPOSE: CALGB (Alliance)/SWOG 80405 was a randomized phase III trial that in first-line patients with metastatic colorectal cancer (mCRC) treated with bevacizumab or cetuximab with chemotherapy. We aimed to discover novel mutated genes associated with prognosis and differential response to therapy with the biologics. METHODS: Primary tumor DNA from 548 patients was sequenced using FoundationOne. The effect of mutated genes and mutations on overall survival (OS) was tested adjusting for microsatellite instability status, BRAF V600E, all RAS mutations, arm, sex, and age. RESULTS: The median number (lower-upper quartile) of mutated genes was 5 (3-7), 5 (3-6) in microsatellite stable and 12.5 (4.5-32) in microsatellite instability-high tumors. Mutated KRAS and APC were more frequent in Black (53% and 85%) than White (27% and 65%, respectively) patients while BRAF V600E was less frequent in Black (5%) than White (14%) patients. The median OS in patients with BRAF non-V600E (2.2% of patients) was 31.9 months (95% CI, 15.1 to not applicable [NA]) similar to that of BRAF wild-type (WT) patients (31.2 months [95% CI, 29.0 to 33.9]). Mutated LRP1B (10.7% of patients) was associated with improved OS compared with WT LRP1B (hazard ratio, 0.57 [95% CI, 0.40 to 0.80]). RNF43 (5.6% of patients) interacted with treatment arms as, in the cetuximab arm, patients with mutated RNF43 had a median OS of 11.5 (95% CI, 10.8 to NA) months compared with 30.1 (95% CI, 24.9 to 35.3) months in patients with WT RNF43, whereas in the bevacizumab arm, patients with mutated RNF43 had a median OS of 25.0 (95% CI, 14.2 to NA) months compared with 31.3 (95% CI, 29.0 to 34.3) months in patients with WT RNF43. CONCLUSION: These results can provide new tools to predict patient outcome and improve therapeutic decisions and trial participation in patient minorities. The molecular alterations identified in this study may direct biomarker-driven studies.

RHOAL57V drives the development of diffuse gastric cancer through IGF1R-PAK1-YAP1 signaling.

Cancer-associated mutations in the guanosine triphosphatase (GTPase) RHOA are found at different locations from the mutational hotspots in the structurally and biochemically related RAS. Tyr42-to-Cys (Y42C) and Leu57-to-Val (L57V) substitutions are the two most prevalent RHOA mutations in diffuse gastric cancer (DGC). RHOAY42C exhibits a gain-of-function phenotype and is an oncogenic driver in DGC. Here, we determined how RHOAL57V promotes DGC growth. In mouse gastric organoids with deletion of Cdh1, which encodes the cell adhesion protein E-cadherin, the expression of RHOAL57V, but not of wild-type RHOA, induced an abnormal morphology similar to that of patient-derived DGC organoids. RHOAL57V also exhibited a gain-of-function phenotype and promoted F-actin stress fiber formation and cell migration. RHOAL57V retained interaction with effectors but exhibited impaired RHOA-intrinsic and GAP-catalyzed GTP hydrolysis, which favored formation of the active GTP-bound state. Introduction of missense mutations at KRAS residues analogous to Tyr42 and Leu57 in RHOA did not activate KRAS oncogenic potential, indicating distinct functional effects in otherwise highly related GTPases. Both RHOA mutants stimulated the transcriptional co-activator YAP1 through actin dynamics to promote DGC progression; however, RHOAL57V additionally did so by activating the kinases IGF1R and PAK1, distinct from the FAK-mediated mechanism induced by RHOAY42C. Our results reveal that RHOAL57V and RHOAY42C drive the development of DGC through distinct biochemical and signaling mechanisms.

Association of PIK3CA Mutation With Pathologic Complete Response and Outcome by Hormone Receptor Status and Intrinsic Subtype in Early-Stage ERBB2/HER2-Positive Breast Cancer.

Importance: PIK3CA mutations may be associated with outcomes of patients with ERBB2/HER2-positive early breast cancer (EBC). Objectives: To assess if PIK3CA mutations among patients with ERBB2/HER2-positive EBC are associated with treatment response and outcome, and if these associations vary by hormone receptor (HR) status or intrinsic molecular subtype (IMS). Design, Setting, and Participants: This cohort study derived data on 184 patients from the phase 3 neoadjuvant Cancer and Leukemia Group B (CALGB) 40601 trial that enrolled patients with ERBB2/HER2-positive EBC in North America between January 1, 2008, and December 31, 2012. Participants received neoadjuvant paclitaxel with trastuzumab, lapatinib, or both. Statistical analysis was performed from March 23, 2022, to March 9, 2023. Exposures: Gene expression profiling by RNA sequencing with Prediction Analysis of Microarray 50-determined IMS and PIK3CA mutations from whole-exome sequencing were obtained from pretreatment biopsies from 184 of 305 trial participants. Main Outcomes and Measures: The primary end point was pathologic complete response (pCR) and the secondary end point of event-free survival (EFS). The association of PIK3CA mutations with pCR and EFS by HR status and IMS was estimated using logistic and Cox proportional hazards regression models. Results: All 184 participants were women, with a median age of 49 years (range 24-75 years). A total of 121 participants (66%) had clinical stage II tumors; 32 (17%) had PIK3CA mutations, most frequently H1047R (38% [12 of 32]) and E545K (22% [7 of 32]). PIK3CA mutations were present in 20 of 102 cases of HR-positive EBC (20%) and 12 of 82 cases HR-negative EBC (15%) and varied by IMS (luminal B, 9 of 25 [36%]; luminal A, 2 of 21 [10%]; and ERBB2/HER2-enriched tumors, 19 of 102 [19%]). Pathologic complete response rates were lower in PIK3CA mutated than PIK3CA wild type in the overall population (34% [11 of 32] vs 49% [74 of 152]; P = .14) and were significantly different among those receiving trastuzumab (30% [7 of 23] vs 54% [63 of 117]; P = .045). At a median follow-up of 9 years, PIK3CA mutations were significantly associated with worse EFS in the overall cohort (hazard ratio, 2.58 [95% CI, 1.24-5.35]; P = .01), which persisted in a multivariable model including pCR, HR status, stage, and IMS (hazard ratio, 2.52 [95% CI, 1.16-5.47]; P = .02). The negative association of PIK3CA mutation was significant in HR-positive (hazard ratio, 3.60 [95% CI, 1.45-8.96]; P = .006) and luminal subtypes (hazard ratio, 4.84 [95% CI, 1.08-21.70]; P = .04), but not in nonluminal and HR-negative tumors. Conclusions and Relevance: In ERBB2/HER2-positive EBC, PIK3CA mutations were associated with lower pCR rates and independently associated with worse long-term EFS. These findings appear to be associated with PIK3CA mutations in HR-positive and luminal EBC.

Sub-Cluster Identification through Semi-Supervised Optimization of Rare-Cell Silhouettes (SCISSORS) in single-cell RNA-sequencing.

MOTIVATION: Single-cell RNA-sequencing (scRNA-seq) has enabled the molecular profiling of thousands to millions of cells simultaneously in biologically heterogenous samples. Currently, the common practice in scRNA-seq is to determine cell type labels through unsupervised clustering and the examination of cluster-specific genes. However, even small differences in analysis and parameter choosing can greatly alter clustering results and thus impose great influence on which cell types are identified. Existing methods largely focus on determining the optimal number of robust clusters, which can be problematic for identifying cells of extremely low abundance due to their subtle contributions toward overall patterns of gene expression. RESULTS: Here, we present a carefully designed framework, SCISSORS, which accurately profiles subclusters within broad cluster(s) for the identification of rare cell types in scRNA-seq data. SCISSORS employs silhouette scoring for the estimation of heterogeneity of clusters and reveals rare cells in heterogenous clusters by a multi-step semi-supervised reclustering process. Additionally, SCISSORS provides a method for the identification of marker genes of high specificity to the cell type. SCISSORS is wrapped around the popular Seurat R package and can be easily integrated into existing Seurat pipelines. AVAILABILITY AND IMPLEMENTATION: SCISSORS, including source code and vignettes, are freely available at https://github.com/jr-leary7/SCISSORS.

Open-source curation of a pancreatic ductal adenocarcinoma gene expression analysis platform (pdacR) supports a two-subtype model.

Pancreatic ductal adenocarcinoma (PDAC) is an aggressive disease for which potent therapies have limited efficacy. Several studies have described the transcriptomic landscape of PDAC tumors to provide insight into potentially actionable gene expression signatures to improve patient outcomes. Despite centralization efforts from multiple organizations and increased transparency requirements from funding agencies and publishers, analysis of public PDAC data remains difficult. Bioinformatic pitfalls litter public transcriptomic data, such as subtle inclusion of low-purity and non-adenocarcinoma cases. These pitfalls can introduce non-specificity to gene signatures without appropriate data curation, which can negatively impact findings. To reduce barriers to analysis, we have created pdacR ( http://pdacR.bmi.stonybrook.edu , github.com/rmoffitt/pdacR), an open-source software package and web-tool with annotated datasets from landmark studies and an interface for user-friendly analysis in clustering, differential expression, survival, and dimensionality reduction. Using this tool, we present a multi-dataset analysis of PDAC transcriptomics that confirms the basal-like/classical model over alternatives.

Prognostic and Predictive Value of Immune-Related Gene Expression Signatures vs Tumor-Infiltrating Lymphocytes in Early-Stage ERBB2/HER2-Positive Breast Cancer: A Correlative Analysis of the CALGB 40601 and PAMELA Trials.

Importance: Both tumor-infiltrating lymphocytes (TILs) assessment and immune-related gene expression signatures by RNA profiling predict higher pathologic complete response (pCR) and improved event-free survival (EFS) in patients with early-stage ERBB2/HER2-positive breast cancer. However, whether these 2 measures of immune activation provide similar or additive prognostic value is not known. Objective: To examine the prognostic ability of TILs and immune-related gene expression signatures, alone and in combination, to predict pCR and EFS in patients with early-stage ERBB2/HER2-positive breast cancer treated in 2 clinical trials. Design, Setting, and Participants: In this prognostic study, a correlative analysis was performed on the Cancer and Leukemia Group B (CALGB) 40601 trial and the PAMELA trial. In the CALGB 40601 trial, 305 patients were randomly assigned to weekly paclitaxel with trastuzumab, lapatinib, or both for 16 weeks. The primary end point was pCR, with a secondary end point of EFS. In the PAMELA trial, 151 patients received neoadjuvant treatment with trastuzumab and lapatinib for 18 weeks. The primary end point was the ability of the HER2-enriched subtype to predict pCR. The studies were conducted from October 2013 to November 2015 (PAMELA) and from December 2008 to February 2012 (CALGB 40601). Data analyses were performed from June 1, 2020, to January 1, 2022. Main Outcomes and Measures: Immune-related gene expression profiling by RNA sequencing and TILs were assessed on 230 CALGB 40601 trial pretreatment tumors and 138 PAMELA trial pretreatment tumors. The association of these biomarkers with pCR (CALGB 40601 and PAMELA) and EFS (CALGB 40601) was studied by logistic regression and Cox analyses. Results: The median age of the patients was 50 years (IQR, 42-50 years), and 305 (100%) were women. Of 202 immune signatures tested, 166 (82.2%) were significantly correlated with TILs. In both trials combined, TILs were significantly associated with pCR (odds ratio, 1.01; 95% CI, 1.01-1.02; P = .02). In addition to TILs, 36 immune signatures were significantly associated with higher pCR rates. Seven of these signatures outperformed TILs for predicting pCR, 6 of which were B-cell related. In a multivariable Cox model adjusted for clinicopathologic factors, including PAM50 intrinsic tumor subtype, the immunoglobulin G signature, but not TILs, was independently associated with EFS (immunoglobulin G signature-adjusted hazard ratio, 0.63; 95% CI, 0.42-0.93; P = .02; TIL-adjusted hazard ratio, 1.00; 95% CI, 0.98-1.02; P = .99). Conclusions and Relevance: Results of this study suggest that multiple B-cell-related signatures were more strongly associated with pCR and EFS than TILs, which largely represent T cells. When both TILs and gene expression are available, the prognostic value of immune-related signatures appears to be superior.

Estimating cell type composition using isoform expression one gene at a time.

Human tissue samples are often mixtures of heterogeneous cell types, which can confound the analyses of gene expression data derived from such tissues. The cell type composition of a tissue sample may itself be of interest and is needed for proper analysis of differential gene expression. A variety of computational methods have been developed to estimate cell type proportions using gene-level expression data. However, RNA isoforms can also be differentially expressed across cell types, and isoform-level expression could be equally or more informative for determining cell type origin than gene-level expression. We propose a new computational method, IsoDeconvMM, which estimates cell type fractions using isoform-level gene expression data. A novel and useful feature of IsoDeconvMM is that it can estimate cell type proportions using only a single gene, though in practice we recommend aggregating estimates of a few dozen genes to obtain more accurate results. We demonstrate the performance of IsoDeconvMM using a unique data set with cell type-specific RNA-seq data across more than 135 individuals. This data set allows us to evaluate different methods given the biological variation of cell type-specific gene expression data across individuals. We further complement this analysis with additional simulations.

glmmPen: High Dimensional Penalized Generalized Linear Mixed Models.

Generalized linear mixed models (GLMMs) are widely used in research for their ability to model correlated outcomes with non-Gaussian conditional distributions. The proper selection of fixed and random effects is a critical part of the modeling process, where model misspecification may lead to significant bias. However, the joint selection of fixed and random effects has historically been limited to lower dimensional GLMMs, largely due to the use of criterion-based model selection strategies. Here we present the R package glmmPen, one of the first to select fixed and random effects in higher dimension using a penalized GLMM modeling framework. Model parameters are estimated using a Monte Carlo expectation conditional minimization (MCECM) algorithm, which leverages Stan and RcppArmadillo for increased computational efficiency. Our package supports the Binomial, Gaussian, and Poisson families and multiple penalty functions. In this manuscript we discuss the modeling procedure, estimation scheme, and software implementation through application to a pancreatic cancer subtyping study. Simulation results show our method has good performance in selecting both the fixed and random effects in high dimensional GLMMs.

Functional and biological heterogeneity of KRASQ61 mutations.

Missense mutations at the three hotspots in the guanosine triphosphatase (GTPase) RAS-Gly12, Gly13, and Gln61 (commonly known as G12, G13, and Q61, respectively)-occur differentially among the three RAS isoforms. Q61 mutations in KRAS are infrequent and differ markedly in occurrence. Q61H is the predominant mutant (at 57%), followed by Q61R/L/K (collectively 40%), and Q61P and Q61E are the rarest (2 and 1%, respectively). Probability analysis suggested that mutational susceptibility to different DNA base changes cannot account for this distribution. Therefore, we investigated whether these frequencies might be explained by differences in the biochemical, structural, and biological properties of KRASQ61 mutants. Expression of KRASQ61 mutants in NIH 3T3 fibroblasts and RIE-1 epithelial cells caused various alterations in morphology, growth transformation, effector signaling, and metabolism. The relatively rare KRASQ61E mutant stimulated actin stress fiber formation, a phenotype distinct from that of KRASQ61H/R/L/P, which disrupted actin cytoskeletal organization. The crystal structure of KRASQ61E was unexpectedly similar to that of wild-type KRAS, a potential basis for its weak oncogenicity. KRASQ61H/L/R-mutant pancreatic ductal adenocarcinoma (PDAC) cell lines exhibited KRAS-dependent growth and, as observed with KRASG12-mutant PDAC, were susceptible to concurrent inhibition of ERK-MAPK signaling and of autophagy. Our results uncover phenotypic heterogeneity among KRASQ61 mutants and support the potential utility of therapeutic strategies that target KRASQ61 mutant-specific signaling and cellular output.

Efficient detection and classification of epigenomic changes under multiple conditions.

Epigenomics, the study of the human genome and its interactions with proteins and other cellular elements, has become of significant interest in recent years. Such interactions have been shown to regulate essential cellular functions and are associated with multiple complex diseases. Therefore, understanding how these interactions may change across conditions is central in biomedical research. Chromatin immunoprecipitation followed by massively parallel sequencing (ChIP-seq) is one of several techniques to detect local changes in epigenomic activity (peaks). However, existing methods for differential peak calling are not optimized for the diversity in ChIP-seq signal profiles, are limited to the analysis of two conditions, or cannot classify specific patterns of differential change when multiple patterns exist. To address these limitations, we present a flexible and efficient method for the detection of differential epigenomic activity across multiple conditions. We utilize data from the ENCODE Consortium and show that the presented method, epigraHMM, exhibits superior performance to current tools and it is among the fastest algorithms available, while allowing the classification of combinatorial patterns of differential epigenomic activity and the characterization of chromatin regulatory states.

The KRAS-regulated kinome identifies WEE1 and ERK coinhibition as a potential therapeutic strategy in KRAS-mutant pancreatic cancer.

Oncogenic KRAS drives cancer growth by activating diverse signaling networks, not all of which have been fully delineated. We set out to establish a system-wide profile of the KRAS-regulated kinase signaling network (kinome) in KRAS-mutant pancreatic ductal adenocarcinoma (PDAC). We knocked down KRAS expression in a panel of six cell lines and then applied multiplexed inhibitor bead/MS to monitor changes in kinase activity and/or expression. We hypothesized that depletion of KRAS would result in downregulation of kinases required for KRAS-mediated transformation and in upregulation of other kinases that could potentially compensate for the deleterious consequences of the loss of KRAS. We identified 15 upregulated and 13 downregulated kinases in common across the panel of cell lines. In agreement with our hypothesis, all 15 of the upregulated kinases have established roles as cancer drivers (e.g., SRC, TGF-ß1, ILK), and pharmacological inhibition of one of these upregulated kinases, DDR1, suppressed PDAC growth. Interestingly, 11 of the 13 downregulated kinases have established driver roles in cell cycle progression, particularly in mitosis (e.g., WEE1, Aurora A, PLK1). Consistent with a crucial role for the downregulated kinases in promoting KRAS-driven proliferation, we found that pharmacological inhibition of WEE1 also suppressed PDAC growth. The unexpected paradoxical activation of ERK upon WEE1 inhibition led us to inhibit both WEE1 and ERK concurrently, which caused further potent growth suppression and enhanced apoptotic death compared with WEE1 inhibition alone. We conclude that system-wide delineation of the KRAS-regulated kinome can identify potential therapeutic targets for KRAS-mutant pancreatic cancer.

High-Dimensional Precision Medicine From Patient-Derived Xenografts.

The complexity of human cancer often results in significant heterogeneity in response to treatment. Precision medicine offers the potential to improve patient outcomes by leveraging this heterogeneity. Individualized treatment rules (ITRs) formalize precision medicine as maps from the patient covariate space into the space of allowable treatments. The optimal ITR is that which maximizes the mean of a clinical outcome in a population of interest. Patient-derived xenograft (PDX) studies permit the evaluation of multiple treatments within a single tumor, and thus are ideally suited for estimating optimal ITRs. PDX data are characterized by correlated outcomes, a high-dimensional feature space, and a large number of treatments. Here we explore machine learning methods for estimating optimal ITRs from PDX data. We analyze data from a large PDX study to identify biomarkers that are informative for developing personalized treatment recommendations in multiple cancers. We estimate optimal ITRs using regression-based (Q-learning) and direct-search methods (outcome weighted learning). Finally, we implement a superlearner approach to combine multiple estimated ITRs and show that the resulting ITR performs better than any of the input ITRs, mitigating uncertainty regarding user choice. Our results indicate that PDX data are a valuable resource for developing individualized treatment strategies in oncology. Supplementary materials for this article are available online.

MODEL-BASED FEATURE SELECTION AND CLUSTERING OF RNA-SEQ DATA FOR UNSUPERVISED SUBTYPE DISCOVERY.

Clustering is a form of unsupervised learning that aims to uncover latent groups within data based on similarity across a set of features. A common application of this in biomedical research is in delineating novel cancer subtypes from patient gene expression data, given a set of informative genes. However, it is typically unknown a priori what genes may be informative in discriminating between clusters, and what the optimal number of clusters are. Few methods exist for performing unsupervised clustering of RNA-seq samples, and none currently adjust for between-sample global normalization factors, select cluster-discriminatory genes, or account for potential confounding variables during clustering. To address these issues, we propose the Feature Selection and Clustering of RNA-seq (FSCseq): a model-based clustering algorithm that utilizes a finite mixture of regression (FMR) model and the quadratic penalty method with a Smoothly-Clipped Absolute Deviation (SCAD) penalty. The maximization is done by a penalized Classification EM algorithm, allowing us to include normalization factors and confounders in our modeling framework. Given the fitted model, our framework allows for subtype prediction in new patients via posterior probabilities of cluster membership, even in the presence of batch effects. Based on simulations and real data analysis, we show the advantages of our method relative to competing approaches.

FOXA1 and adaptive response determinants to HER2 targeted therapy in TBCRC 036.

Inhibition of the HER2/ERBB2 receptor is a keystone to treating HER2-positive malignancies, particularly breast cancer, but a significant fraction of HER2-positive (HER2+) breast cancers recur or fail to respond. Anti-HER2 monoclonal antibodies, like trastuzumab or pertuzumab, and ATP active site inhibitors like lapatinib, commonly lack durability because of adaptive changes in the tumor leading to resistance. HER2+ cell line responses to inhibition with lapatinib were analyzed by RNAseq and ChIPseq to characterize transcriptional and epigenetic changes. Motif analysis of lapatinib-responsive genomic regions implicated the pioneer transcription factor FOXA1 as a mediator of adaptive responses. Lapatinib in combination with FOXA1 depletion led to dysregulation of enhancers, impaired adaptive upregulation of HER3, and decreased proliferation. HER2-directed therapy using clinically relevant drugs (trastuzumab with or without lapatinib or pertuzumab) in a 7-day clinical trial designed to examine early pharmacodynamic response to antibody-based anti-HER2 therapy showed reduced FOXA1 expression was coincident with decreased HER2 and HER3 levels, decreased proliferation gene signatures, and increased immune gene signatures. This highlights the importance of the immune response to anti-HER2 antibodies and suggests that inhibiting FOXA1-mediated adaptive responses in combination with HER2 targeting is a potential therapeutic strategy.

Compression of quantification uncertainty for scRNA-seq counts.

MOTIVATION: Quantification estimates of gene expression from single-cell RNA-seq (scRNA-seq) data have inherent uncertainty due to reads that map to multiple genes. Many existing scRNA-seq quantification pipelines ignore multi-mapping reads and therefore underestimate expected read counts for many genes. alevin accounts for multi-mapping reads and allows for the generation of 'inferential replicates', which reflect quantification uncertainty. Previous methods have shown improved performance when incorporating these replicates into statistical analyses, but storage and use of these replicates increases computation time and memory requirements. RESULTS: We demonstrate that storing only the mean and variance from a set of inferential replicates ('compression') is sufficient to capture gene-level quantification uncertainty, while reducing disk storage to as low as 9% of original storage, and memory usage when loading data to as low as 6%. Using these values, we generate 'pseudo-inferential' replicates from a negative binomial distribution and propose a general procedure for incorporating these replicates into a proposed statistical testing framework. When applying this procedure to trajectory-based differential expression analyses, we show false positives are reduced by more than a third for genes with high levels of quantification uncertainty. We additionally extend the Swish method to incorporate pseudo-inferential replicates and demonstrate improvements in computation time and memory usage without any loss in performance. Lastly, we show that discarding multi-mapping reads can result in significant underestimation of counts for functionally important genes in a real dataset. AVAILABILITY AND IMPLEMENTATION: makeInfReps and splitSwish are implemented in the R/Bioconductor fishpond package available at https://bioconductor.org/packages/fishpond. Analyses and simulated datasets can be found in the paper's GitHub repo at https://github.com/skvanburen/scUncertaintyPaperCode. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Modeling Between-Study Heterogeneity for Improved Replicability in Gene Signature Selection and Clinical Prediction.

In the genomic era, the identification of gene signatures associated with disease is of significant interest. Such signatures are often used to predict clinical outcomes in new patients and aid clinical decision-making. However, recent studies have shown that gene signatures are often not replicable. This occurrence has practical implications regarding the generalizability and clinical applicability of such signatures. To improve replicability, we introduce a novel approach to select gene signatures from multiple datasets whose effects are consistently non-zero and account for between-study heterogeneity. We build our model upon some rank-based quantities, facilitating integration over different genomic datasets. A high dimensional penalized Generalized Linear Mixed Model (pGLMM) is used to select gene signatures and address data heterogeneity. We compare our method to some commonly used strategies that select gene signatures ignoring between-study heterogeneity. We provide asymptotic results justifying the performance of our method and demonstrate its advantage in the presence of heterogeneity through thorough simulation studies. Lastly, we motivate our method through a case study subtyping pancreatic cancer patients from four gene expression studies.