Machine Learning Prediction of Lymph Node Metastasis in Breast Cancer: Performance of a Multi-institutional MRI-based 4D Convolutional Neural Network.

Purpose To develop a custom deep convolutional neural network (CNN) for noninvasive prediction of breast cancer nodal metastasis. Materials and Methods This retrospective study included patients with newly diagnosed primary invasive breast cancer with known pathologic (pN) and clinical nodal (cN) status who underwent dynamic contrast-enhanced (DCE) breast MRI at the authors' institution between July 2013 and July 2016. Clinicopathologic data (age, estrogen receptor and human epidermal growth factor 2 status, Ki-67 index, and tumor grade) and cN and pN status were collected. A four-dimensional (4D) CNN model integrating temporal information from dynamic image sets was developed. The convolutional layers learned prognostic image features, which were combined with clinicopathologic measures to predict cN0 versus cN+ and pN0 versus pN+ disease. Performance was assessed with the area under the receiver operating characteristic curve (AUC), with fivefold nested cross-validation. Results Data from 350 female patients (mean age, 51.7 years ± 11.9 [SD]) were analyzed. AUC, sensitivity, and specificity values of the 4D hybrid model were 0.87 (95% CI: 0.83, 0.91), 89% (95% CI: 79%, 93%), and 76% (95% CI: 68%, 88%) for differentiating pN0 versus pN+ and 0.79 (95% CI: 0.76, 0.82), 80% (95% CI: 77%, 84%), and 62% (95% CI: 58%, 67%), respectively, for differentiating cN0 versus cN+. Conclusion The proposed deep learning model using tumor DCE MR images demonstrated high sensitivity in identifying breast cancer lymph node metastasis and shows promise for potential use as a clinical decision support tool. Keywords: MR Imaging, Breast, Breast Cancer, Breast MRI, Machine Learning, Metastasis, Prognostic Prediction Supplemental material is available for this article. Published under a CC BY 4.0 license.

Adrenergic signaling controls early transcriptional programs during CD8+ T cell responses to viral infection.

Norepinephrine is a key sympathetic neurotransmitter, which acts to suppress CD8 + T cell cytokine secretion and lytic activity by signaling through the ß2-adrenergic receptor (ADRB2). Although ADRB2 signaling is considered generally immunosuppressive, its role in regulating the differentiation of effector T cells in response to infection has not been investigated. Using an adoptive transfer approach, we compared the expansion and differentiation of wild type (WT) to Adrb2-/- CD8 + T cells throughout the primary response to vesicular stomatitis virus (VSV) infection in vivo. We measured the dynamic changes in transcriptome profiles of antigen-specific CD8 + T cells as they responded to VSV. Within the first 7 days of infection, WT cells out-paced the expansion of Adrb2-/- cells, which correlated with reduced expression of IL-2 and the IL-2Rα in the absence of ADRB2. RNASeq analysis identified over 300 differentially expressed genes that were both temporally regulated following infection and selectively regulated in WT vs Adrb2-/- cells. These genes contributed to major transcriptional pathways including cytokine receptor activation, signaling in cancer, immune deficiency, and neurotransmitter pathways. By parsing genes within groups that were either induced or repressed over time in response to infection, we identified three main branches of genes that were differentially regulated by the ADRB2. These gene sets were predicted to be regulated by specific transcription factors involved in effector T cell development, such as Tbx21 and Eomes. Collectively, these data demonstrate a significant role for ADRB2 signaling in regulating key transcriptional pathways during CD8 + T cells responses to infection that may dramatically impact their functional capabilities and downstream memory cell development.

Haisu: Hierarchically supervised nonlinear dimensionality reduction.

We propose a novel strategy for incorporating hierarchical supervised label information into nonlinear dimensionality reduction techniques. Specifically, we extend t-SNE, UMAP, and PHATE to include known or predicted class labels and demonstrate the efficacy of our approach on multiple single-cell RNA sequencing datasets. Our approach, "Haisu," is applicable across domains and methods of nonlinear dimensionality reduction. In general, the mathematical effect of Haisu can be summarized as a variable perturbation of the high dimensional space in which the original data is observed. We thereby preserve the core characteristics of the visualization method and only change the manifold to respect known or assumed class labels when provided. Our strategy is designed to aid in the discovery and understanding of underlying patterns in a dataset that is heavily influenced by parent-child relationships. We show that using our approach can also help in semi-supervised settings where labels are known for only some datapoints (for instance when only a fraction of the cells are labeled). In summary, Haisu extends existing popular visualization methods to enable a user to incorporate labels known a priori into a visualization, including their hierarchical relationships as defined by a user input graph.

Evaluating Melanoma Viability and Proliferation in 3D Microenvironments.

Researchers often aim to incorporate microenvironmental variables such as the dimensionality and composition of the extracellular matrix into their cell-based assays. A technical challenge created by introduction of these variables is quantification of single-cell measurements and control of environmental reproducibility. Here, we detail a methodology to quantify viability and proliferation of melanoma cells in 3D collagen-based culture platforms by automated microscopy and 3D image analysis to yield robust, high-throughput results of single-cell responses to drug treatment.

Preoperative Prediction of Lymph Node Metastasis from Clinical DCE MRI of the Primary Breast Tumor Using a 4D CNN.

In breast cancer, undetected lymph node metastases can spread to distal parts of the body for which the 5-year survival rate is only 27%, making accurate nodal metastases diagnosis fundamental to reducing the burden of breast cancer, when it is still early enough to intervene with surgery and adjuvant therapies. Currently, breast cancer management entails a time consuming and costly sequence of steps to clinically diagnose axillary nodal metastases status. The purpose of this study is to determine whether preoperative, clinical DCE MRI of the primary tumor alone may be used to predict clinical node status with a deep learning model. If possible then many costly steps could be eliminated or reserved for only those with uncertain or probable nodal metastases. This research develops a data-driven approach that predicts lymph node metastasis through the judicious integration of clinical and imaging features from preoperative 4D dynamic contrast enhanced (DCE) MRI of 357 patients from 2 hospitals. Innovative deep learning classifiers are trained from scratch, including 2D, 3D, 4D and 4D deep convolutional neural networks (CNNs) that integrate multiple data types and predict the nodal metastasis differentiating nodal stage N0 (non metastatic) against stages N1, N2 and N3. Appropriate methodologies for data preprocessing and network interpretation are presented, the later of which bolster radiologist confidence that the model has learned relevant features from the primary tumor. Rigorous nested 10-fold cross-validation provides an unbiased estimate of model performance. The best model achieves a high sensitivity of 72% and an AUROC of 71% on held out test data. Results are strongly supportive of the potential of the combination of DCE MRI and machine learning to inform diagnostics that could substantially reduce breast cancer burden.

An image-based assay to quantify changes in proliferation and viability upon drug treatment in 3D microenvironments.

BACKGROUND: Every biological experiment requires a choice of throughput balanced against physiological relevance. Most primary drug screens neglect critical parameters such as microenvironmental conditions, cell-cell heterogeneity, and specific readouts of cell fate for the sake of throughput. METHODS: Here we describe a methodology to quantify proliferation and viability of single cells in 3D culture conditions by leveraging automated microscopy and image analysis to facilitate reliable and high-throughput measurements. We detail experimental conditions that can be adjusted to increase either throughput or robustness of the assay, and we provide a stand alone image analysis program for users who wish to implement this 3D drug screening assay in high throughput. RESULTS: We demonstrate this approach by evaluating a combination of RAF and MEK inhibitors on melanoma cells, showing that cells cultured in 3D collagen-based matrices are more sensitive than cells grown in 2D culture, and that cell proliferation is much more sensitive than cell viability. We also find that cells grown in 3D cultured spheroids exhibit equivalent sensitivity to single cells grown in 3D collagen, suggesting that for the case of melanoma, a 3D single cell model may be equally effective for drug identification as 3D spheroids models. The single cell resolution of this approach enables stratification of heterogeneous populations of cells into differentially responsive subtypes upon drug treatment, which we demonstrate by determining the effect of RAK/MEK inhibition on melanoma cells co-cultured with fibroblasts. Furthermore, we show that spheroids grown from single cells exhibit dramatic heterogeneity to drug response, suggesting that heritable drug resistance can arise stochastically in single cells but be retained by subsequent generations. CONCLUSION: In summary, image-based analysis renders cell fate detection robust, sensitive, and high-throughput, enabling cell fate evaluation of single cells in more complex microenvironmental conditions.

Accurate differential analysis of transcription factor activity from gene expression.

MOTIVATION: Activity of transcriptional regulators is crucial in elucidating the mechanism of phenotypes. However regulatory activity hypotheses are difficult to experimentally test. Therefore, we need accurate and reliable computational methods for regulator activity inference. There is extensive work in this area, however, current methods have difficulty with one or more of the following: resolving activity of TFs with overlapping regulons, reflecting known regulatory relationships, or flexible modeling of TF activity over the regulon. RESULTS: We present Effector and Perturbation Estimation Engine (EPEE), a method for differential analysis of transcription factor (TF) activity from gene expression data. EPEE addresses each of these principal challenges in the field. Firstly, EPEE collectively models all TF activity in a single multivariate model, thereby accounting for the intrinsic coupling among TFs that share targets, which is highly frequent. Secondly, EPEE incorporates context-specific TF-gene regulatory networks and therefore adapts the analysis to each biological context. Finally, EPEE can flexibly reflect different regulatory activity of a single TF among its potential targets. This allows the flexibility to implicitly recover other regulatory influences such as co-activators or repressors. We comparatively validated EPEE in 15 datasets from three well-studied contexts, namely immunology, cancer, and hematopoiesis. We show that addressing the aforementioned challenges enable EPEE to outperform alternative methods and reliably produce accurate results. AVAILABILITY AND IMPLEMENTATION: https://github.com/Cobanoglu-Lab/EPEE. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Connecting Neuronal Cell Protective Pathways and Drug Combinations in a Huntington's Disease Model through the Application of Quantitative Systems Pharmacology.

Quantitative Systems Pharmacology (QSP) is a drug discovery approach that integrates computational and experimental methods in an iterative way to gain a comprehensive, unbiased understanding of disease processes to inform effective therapeutic strategies. We report the implementation of QSP to Huntington's Disease, with the application of a chemogenomics platform to identify strategies to protect neuronal cells from mutant huntingtin induced death. Using the STHdh Q111 cell model, we investigated the protective effects of small molecule probes having diverse canonical modes-of-action to infer pathways of neuronal cell protection connected to drug mechanism. Several mechanistically diverse protective probes were identified, most of which showed less than 50% efficacy. Specific combinations of these probes were synergistic in enhancing efficacy. Computational analysis of these probes revealed a convergence of pathways indicating activation of PKA. Analysis of phospho-PKA levels showed lower cytoplasmic levels in STHdh Q111 cells compared to wild type STHdh Q7 cells, and these levels were increased by several of the protective compounds. Pharmacological inhibition of PKA activity reduced protection supporting the hypothesis that protection may be working, in part, through activation of the PKA network. The systems-level studies described here can be broadly applied to any discovery strategy involving small molecule modulation of disease phenotype.

Insights into the Modulation of Dopamine Transporter Function by Amphetamine, Orphenadrine, and Cocaine Binding.

Human dopamine (DA) transporter (hDAT) regulates dopaminergic signaling in the central nervous system by maintaining the synaptic concentration of DA at physiological levels, upon reuptake of DA into presynaptic terminals. DA translocation involves the co-transport of two sodium ions and the channeling of a chloride ion, and it is achieved via alternating access between outward-facing (OF) and inward-facing states of DAT. hDAT is a target for addictive drugs, such as cocaine, amphetamine (AMPH), and therapeutic antidepressants. Our recent quantitative systems pharmacology study suggested that orphenadrine (ORPH), an anticholinergic agent and anti-Parkinson drug, might be repurposable as a DAT drug. Previous studies have shown that DAT-substrates like AMPH or -blockers like cocaine modulate the function of DAT in different ways. However, the molecular mechanisms of modulation remained elusive due to the lack of structural data on DAT. The newly resolved DAT structure from Drosophila melanogaster opens the way to a deeper understanding of the mechanism and time evolution of DAT-drug/ligand interactions. Using a combination of homology modeling, docking analysis, molecular dynamics simulations, and molecular biology experiments, we performed a comparative study of the binding properties of DA, AMPH, ORPH, and cocaine and their modulation of hDAT function. Simulations demonstrate that binding DA or AMPH drives a structural transition toward a functional form predisposed to translocate the ligand. In contrast, ORPH appears to inhibit DAT function by arresting it in the OF open conformation. The analysis shows that cocaine and ORPH competitively bind DAT, with the binding pose and affinity dependent on the conformational state of DAT. Further assays show that the effect of ORPH on DAT uptake and endocytosis is comparable to that of cocaine.

BalestraWeb: efficient online evaluation of drug-target interactions.

SUMMARY: BalestraWeb is an online server that allows users to instantly make predictions about the potential occurrence of interactions between any given drug-target pair, or predict the most likely interaction partners of any drug or target listed in the DrugBank. It also permits users to identify most similar drugs or most similar targets based on their interaction patterns. Outputs help to develop hypotheses about drug repurposing as well as potential side effects. AVAILABILITY AND IMPLEMENTATION: BalestraWeb is accessible at http://balestra.csb.pitt.edu/. The tool is built using a probabilistic matrix factorization method and DrugBank v3, and the latent variable models are trained using the GraphLab collaborative filtering toolkit. The server is implemented using Python, Flask, NumPy and SciPy.

Predicting drug-target interactions using probabilistic matrix factorization.

Quantitative analysis of known drug-target interactions emerged in recent years as a useful approach for drug repurposing and assessing side effects. In the present study, we present a method that uses probabilistic matrix factorization (PMF) for this purpose, which is particularly useful for analyzing large interaction networks. DrugBank drugs clustered based on PMF latent variables show phenotypic similarity even in the absence of 3D shape similarity. Benchmarking computations show that the method outperforms those recently introduced provided that the input data set of known interactions is sufficiently large--which is the case for enzymes and ion channels, but not for G-protein coupled receptors (GPCRs) and nuclear receptors. Runs performed on DrugBank after hiding 70% of known interactions show that, on average, 88 of the top 100 predictions hit the hidden interactions. De novo predictions permit us to identify new potential interactions. Drug-target pairs implicated in neurobiological disorders are overrepresented among de novo predictions.

Ionizing irradiation protection and mitigation of murine cells by carbamazepine is p53 and autophagy independent.

BACKGROUND: Carbamazepine, a sodium channel blocker and pro-autophagy agent used in the treatment of epilepsy and trigeminal neuralgia, is also an ionizing radiation mitigator and protector. MATERIALS AND METHODS: We measured the effect of carbamazepine, compared to other pro-autophagy drugs (i.e. lithium and valproic acid), on irradiation of autophagy incompetent (Atg5(-/-)) and competent (Atg5(+/+)) mouse embryonic fibroblasts, p53(-/-) and p53(+/+) bone marrow stromal cells, and human IB3, KM101, HeLa, and umbilical cord blood cell and in total body-irradiated or orthotopic tumor-bearing mice. RESULTS: Carbamazepine, but not other pro-autophagy drugs, was a radiation protector and mitigator for mouse cell lines, independent of apoptosis, autophagy, p53, antioxidant store depletion, and class I phosphatidylinositol 3-kinase, but was ineffective with human cells. Carbamazepine was effective when delivered 24 hours before or 12 hours after total body irradiation of C57BL/6HNsd mice and did not protect orthotopic Lewis lung tumors. CONCLUSION: Carbamazepine is a murine radiation protector and mitigator.

Classification of GPCRs using family specific motifs.

The classification of G-Protein Coupled Receptor (GPCR) sequences is an important problem that arises from the need to close the gap between the large number of orphan receptors and the relatively small number of annotated receptors. Equally important is the characterization of GPCR Class A subfamilies and gaining insight into the ligand interaction since GPCR Class A encompasses a very large number of drug-targeted receptors. In this work, we propose a method for Class A subfamily classification using sequence-derived motifs which characterizes the subfamilies by discovering receptor-ligand interaction sites. The motifs that best characterize a subfamily are selected by the Distinguishing Power Evaluation (DPE) technique we propose. The experiments performed on GPCR sequence databases show that our method outperforms state-of-the-art classification techniques for GPCR Class A subfamily prediction. An important contribution of our work is to discover key receptor-ligand interaction sites which is very important for drug design.