Development of a multigenomic liquid biopsy (PROSTest) for prostate cancer in whole blood.

INTRODUCTION: We describe the development of a molecular assay from publicly available tumor tissue mRNA databases using machine learning and present preliminary evidence of functionality as a diagnostic and monitoring tool for prostate cancer (PCa) in whole blood. MATERIALS AND METHODS: We assessed 1055 PCas (public microarray data sets) to identify putative mRNA biomarkers. Specificity was confirmed against 32 different solid and hematological cancers from The Cancer Genome Atlas (n = 10,990). This defined a 27-gene panel which was validated by qPCR in 50 histologically confirmed PCa surgical specimens and matched blood. An ensemble classifier (Random Forest, Support Vector Machines, XGBoost) was trained in age-matched PCas (n = 294), and in 72 controls and 64 BPH. Classifier performance was validated in two independent sets (n = 263 PCas; n = 99 controls). We assessed the panel as a postoperative disease monitor in a radical prostatectomy cohort (RPC: n = 47). RESULTS: A PCa-specific 27-gene panel was identified. Matched blood and tumor gene expression levels were concordant (r = 0.72, p < 0.0001). The ensemble classifier ("PROSTest") was scaled 0%-100% and the industry-standard operating point of ≥50% used to define a PCa. Using this, the PROSTest exhibited an 85% sensitivity and 95% specificity for PCa versus controls. In two independent sets, the metrics were 92%-95% sensitivity and 100% specificity. In the RPCs (n = 47), PROSTest scores decreased from 72% ± 7% to 33% ± 16% (p < 0.0001, Mann-Whitney test). PROSTest was 26% ± 8% in 37 with normal postoperative PSA levels (<0.1 ng/mL). In 10 with elevated postoperative PSA, PROSTest was 60% ± 4%. CONCLUSION: A 27-gene whole blood signature for PCa is concordant with tissue mRNA levels. Measuring blood expression provides a minimally invasive genomic tool that may facilitate prostate cancer management.

A data-driven approach to decode metabolic dysfunction-associated steatotic liver disease.

Metabolic dysfunction-associated steatotic liver disease (MASLD), defined by the presence of liver steatosis together with at least one out of five cardiometabolic factors, is the most common cause of chronic liver disease worldwide, affecting around one in three people. Yet the clinical presentation of MASLD and the risk of progression to cirrhosis and adverse clinical outcomes is highly variable. It, therefore, represents both a global public health threat and a precision medicine challenge. Artificial intelligence (AI) is being investigated in MASLD to develop reproducible, quantitative, and automated methods to enhance patient stratification and to discover new biomarkers and therapeutic targets in MASLD. This review details the different applications of AI and machine learning algorithms in MASLD, particularly in analyzing electronic health record, digital pathology, and imaging data. Additionally, it also describes how specific MASLD consortia are leveraging multimodal data sources to spark research breakthroughs in the field. Using a new national-level 'data commons' (SteatoSITE) as an exemplar, the opportunities, as well as the technical challenges of large-scale databases in MASLD research, are highlighted.

An Artificial Neural Network for Nasogastric Tube Position Decision Support.

Purpose: To develop and validate a deep learning model for detection of nasogastric tube (NGT) malposition on chest radiographs and assess model impact as a clinical decision support tool for junior physicians to help determine whether feeding can be safely performed in patients (feed/do not feed). Materials and Methods: A neural network ensemble was pretrained on 1 132 142 retrospectively collected (June 2007-August 2019) frontal chest radiographs and further fine-tuned on 7081 chest radiographs labeled by three radiologists. Clinical relevance was assessed on an independent set of 335 images. Five junior emergency medicine physicians assessed chest radiographs and made feed/do not feed decisions without and with artificial intelligence (AI)-generated NGT malposition probabilities placed above chest radiographs. Decisions from the radiologists served as ground truths. Model performance was evaluated using receiver operating characteristic analysis. Agreement between junior physician and radiologist decision was determined using the Cohen κ coefficient. Results: In the testing set, the ensemble achieved area under the receiver operating characteristic curve values of 0.82 (95% CI: 0.78, 0.86), 0.77 (95% CI: 0.71, 0.83), and 0.98 (95% CI: 0.96, 1.00) for satisfactory, malpositioned, and bronchial positions, respectively. In the clinical evaluation set, mean interreader agreement for feed/do not feed decisions among junior physicians was 0.65 ± 0.03 (SD) and 0.77 ± 0.13 without and with AI support, respectively. Mean agreement between junior physicians and radiologists was 0.53 ± 0.05 (unaided) and 0.65 ± 0.09 (AI-aided). Conclusion: A simple classifier for NGT malposition may help junior physicians determine the safety of feeding in patients with NGTs.Keywords: Neural Networks, Feature Detection, Supervised Learning, Machine Learning Supplemental material is available for this article. Published under a CC BY 4.0 license.

C-type lectin receptor CLEC4A2 promotes tissue adaptation of macrophages and protects against atherosclerosis.

Macrophages are integral to the pathogenesis of atherosclerosis, but the contribution of distinct macrophage subsets to disease remains poorly defined. Using single cell technologies and conditional ablation via a LysMCre+ Clec4a2flox/DTR mouse strain, we demonstrate that the expression of the C-type lectin receptor CLEC4A2 is a distinguishing feature of vascular resident macrophages endowed with athero-protective properties. Through genetic deletion and competitive bone marrow chimera experiments, we identify CLEC4A2 as an intrinsic regulator of macrophage tissue adaptation by promoting a bias in monocyte-to-macrophage in situ differentiation towards colony stimulating factor 1 (CSF1) in vascular health and disease. During atherogenesis, CLEC4A2 deficiency results in loss of resident vascular macrophages and their homeostatic properties causing dysfunctional cholesterol metabolism and enhanced toll-like receptor triggering, exacerbating disease. Our study demonstrates that CLEC4A2 licenses monocytes to join the vascular resident macrophage pool, and that CLEC4A2-mediated macrophage homeostasis is critical to combat cardiovascular disease.

Development and prospective validation of COVID-19 chest X-ray screening model for patients attending emergency departments.

Chest X-rays (CXRs) are the first-line investigation in patients presenting to emergency departments (EDs) with dyspnoea and are a valuable adjunct to clinical management of COVID-19 associated lung disease. Artificial intelligence (AI) has the potential to facilitate rapid triage of CXRs for further patient testing and/or isolation. In this work we develop an AI algorithm, CovIx, to differentiate normal, abnormal, non-COVID-19 pneumonia, and COVID-19 CXRs using a multicentre cohort of 293,143 CXRs. The algorithm is prospectively validated in 3289 CXRs acquired from patients presenting to ED with symptoms of COVID-19 across four sites in NHS Greater Glasgow and Clyde. CovIx achieves area under receiver operating characteristic curve for COVID-19 of 0.86, with sensitivity and F1-score up to 0.83 and 0.71 respectively, and performs on-par with four board-certified radiologists. AI-based algorithms can identify CXRs with COVID-19 associated pneumonia, as well as distinguish non-COVID pneumonias in symptomatic patients presenting to ED. Pre-trained models and inference scripts are freely available at https://github.com/beringresearch/bravecx-covid .

Early Identification of Residual Disease After Neuroendocrine Tumor Resection Using a Liquid Biopsy Multigenomic mRNA Signature (NETest).

INTRODUCTION: Surgery is the only cure for neuroendocrine tumors (NETs), with R0 resection being critical for successful tumor removal. Early detection of residual disease is key for optimal management, but both imaging and current biomarkers are ineffective post-surgery. NETest, a multigene blood biomarker, identifies NETs with >90% accuracy. We hypothesized that surgery would decrease NETest levels and that elevated scores post-surgery would predict recurrence. METHODS: This was a multicenter evaluation of surgically treated primary NETs (n = 153). Blood sampling was performed at day 0 and postoperative day (POD) 30. Follow-up included computed tomography/magnetic resonance imaging (CT/MRI), and messenger RNA (mRNA) quantification was performed by polymerase chain reaction (PCR; NETest score: 0-100; normal ≤20). Statistical analyses were performed using the Mann-Whitney U-test, Chi-square test, Kaplan-Meier survival, and area under the receiver operating characteristic curve (AUROC), as appropriate. Data are presented as mean ± standard deviation. RESULTS: The NET cohort (n = 153) included 57 patients with pancreatic cancer, 62 patients with small bowel cancer, 27 patients with lung cancer, 4 patients with duodenal cancer, and 3 patients with gastric cancer, while the surgical cohort comprised patients with R0 (n = 102) and R1 and R2 (n = 51) resection. The mean follow-up time was 14 months (range 3-68). The NETest was positive in 153/153 (100%) samples preoperatively (mean levels of 68 ± 28). In the R0 cohort, POD30 levels decreased from 62 ± 28 to 22 ± 20 (p < 0.0001), but remained elevated in 30% (31/102) of patients: 28% lung, 29% pancreas, 27% small bowel, and 33% gastric. By 18 months, 25/31 (81%) patients with a POD30 NETest >20 had image-identifiable recurrence. An NETest score of >20 predicted recurrence with 100% sensitivity and correlated with residual disease (Chi-square 17.1, p < 0.0001). AUROC analysis identified an AUC of 0.97 (p < 0.0001) for recurrence-prediction. In the R1 (n = 29) and R2 (n = 22) cohorts, the score decreased (R1: 74 ± 28 to 45 ± 24, p = 0.0012; R2: 72 ± 24 to 60 ± 28, p = non-significant). At POD30, 100% of NETest scores were elevated despite surgery (p < 0.0001). CONCLUSION: The preoperative NETest accurately identified all NETs (100%). All resections decreased NETest levels and a POD30 NETest score >20 predicted radiologically recurrent disease with 94% accuracy and 100% sensitivity. R0 resection appears to be ineffective in approximately 30% of patients. NET mRNA blood levels provide early objective genomic identification of residual disease and may facilitate management.

Neuroendocrine Tumor Omic Gene Cluster Analysis Amplifies the Prognostic Accuracy of the NETest.

BACKGROUND: The NETest is a multigene assay comprising 51 circulating neuroendocrine tumor (NET)-specific transcripts. The quotient of the 51-gene assay is based upon an ensemble of machine learning algorithms. Eight cancer hallmarks or "omes" (apoptome, epigenome, growth factor signalome, metabolome, proliferome, plurome, secretome, SSTRome) represent 29 genes. The NETest is an accurate diagnostic (>90%) test, but its prognostic utility has not been assessed. In this study, we describe the expansion of the NETest omic cluster components and demonstrate that integration amplifies NETest prognostic accuracy. METHODS: Group 1: n = 222; including stable disease (SD, n = 146), progressive disease (PD, n = 76), and controls (n = 139). Group 2: NET Registry NCT02270567; n = 88; prospective samples (SD, n = 54; PD, n = 34) with up to 24 months follow-up. We used PubMed literature review, interactomic analysis, nonparametric testing, Kaplan-Meier survival curves, and χ2 analyses to inform and define the prognostic significance of NET genomic "hallmarks." RESULTS: 2020 analyses: In-depth analyses of 47 -NETest genes identified a further six omes: fibrosome, inflammasome, metastasome, NEDome, neurome, and TFome. Group 1 analysis: Twelve omes, excluding the inflammasome and apoptome, were significantly (p < 0.05, 2.1- to 8.2-fold) elevated compared to controls. In the PD group, seven omes (proliferome, NEDome, epigenome, SSTRome, neurome, metastasome, and fibrosome) were elevated (both expression levels and fold change >2) versus SD. Group 2 analysis: All these seven omes were upregulated. In PD, they were significantly more elevated (p < 0.02) than in SD. The septet omic expression exhibited a 69% prognostic accuracy. The NETest alone was 70.5% accurate. A low NETest (≤40) integrated with epigenome/metastasome levels was an accurate prognostic for PD (90%). A high NETest (>40) including the fibrosome/NEDome predicted PD development within 3 months (100%). Using decision tree analysis to integrate the four omes (epigenome, metastasome, fibrosome, and NEDome) with the NETest score generated an overall prognostic accuracy of 93%. CONCLUSIONS: Examination of NETest omic gene cluster analysis identified five additional clinically relevant cancer hallmarks. Identification of seven omic clusters (septet) provides a molecular pathological signature of disease progression. The integration of the quartet (epigenome, fibrosome, metastasome, NEDome) and the NETest score yielded a 93% accuracy in the prediction of future disease status.

Supervised and unsupervised language modelling in Chest X-Ray radiological reports.

Chest radiography (CXR) is the most commonly used imaging modality and deep neural network (DNN) algorithms have shown promise in effective triage of normal and abnormal radiograms. Typically, DNNs require large quantities of expertly labelled training exemplars, which in clinical contexts is a major bottleneck to effective modelling, as both considerable clinical skill and time is required to produce high-quality ground truths. In this work we evaluate thirteen supervised classifiers using two large free-text corpora and demonstrate that bi-directional long short-term memory (BiLSTM) networks with attention mechanism effectively identify Normal, Abnormal, and Unclear CXR reports in internal (n = 965 manually-labelled reports, f1-score = 0.94) and external (n = 465 manually-labelled reports, f1-score = 0.90) testing sets using a relatively small number of expert-labelled training observations (n = 3,856 annotated reports). Furthermore, we introduce a general unsupervised approach that accurately distinguishes Normal and Abnormal CXR reports in a large unlabelled corpus. We anticipate that the results presented in this work can be used to automatically extract standardized clinical information from free-text CXR radiological reports, facilitating the training of clinical decision support systems for CXR triage.

Microbiota-Derived Metabolites Suppress Arthritis by Amplifying Aryl-Hydrocarbon Receptor Activation in Regulatory B Cells.

The differentiation of IL-10-producing regulatory B cells (Bregs) in response to gut-microbiota-derived signals supports the maintenance of tolerance. However, whether microbiota-derived metabolites can modulate Breg suppressive function remains unknown. Here, we demonstrate that rheumatoid arthritis (RA) patients and arthritic mice have a reduction in microbial-derived short-chain fatty acids (SCFAs) compared to healthy controls and that in mice, supplementation with the SCFA butyrate reduces arthritis severity. Butyrate supplementation suppresses arthritis in a Breg-dependent manner by increasing the level of the serotonin-derived metabolite 5-Hydroxyindole-3-acetic acid (5-HIAA), which activates the aryl-hydrocarbon receptor (AhR), a newly discovered transcriptional marker for Breg function. Thus, butyrate supplementation via AhR activation controls a molecular program that supports Breg function while inhibiting germinal center (GC) B cell and plasmablast differentiation. Our study demonstrates that butyrate supplementation may serve as a viable therapy for the amelioration of systemic autoimmune disorders.

Blood Chromogranin A Is Not Effective as a Biomarker for Diagnosis or Management of Bronchopulmonary Neuroendocrine Tumors/Neoplasms.

BACKGROUND: Identification of circulating tumor markers for clinical management in bronchopulmonary (BP) neuroendocrine tumors/neoplasms (NET/NEN) is of considerable clinical interest. Chromogranin A (CgA), a "universal" NET biomarker, is considered controversial as a circulating biomarker of BPNEN. AIM: Assess utility of CgA in the diagnosis and management of BPNEN in a multicentric study. MATERIAL AND METHODS: CgA diagnostic metrics were assessed in lung NET/NENs (n = 200) and controls (n = 140), randomly assigned to a Training and Test set (100 BPC and 70 controls in each). Assay specificity was evaluated in neoplastic lung disease (n = 137) and nonneoplastic lung disease (n = 77). CgA efficacy in predicting clinical status was evaluated in the combined set of 200 NET/NENs. CgA levels in bronchopulmonary neuroendocrine tumor (BPNET) subtypes (atypical [AC] vs. typical [TC]) and grade was examined. The clinical utility of an alteration of CgA levels (±25%) was evaluated in a subset of 49 BPNET over 12 months. CgA measurement was by NEOLISATM kit (EuroDiagnostica). RESULTS: Sensitivity and specificity in the training set were 41/98%, respectively. Test set data were 42/87%. Training set area under receiver operator characteristic analysis differentiated BPC from control area under the curve (AUC) 0.61 ± 0.05 p = 0.015. Test set the data were AUC 0.58 ± 0.05, p = 0.076. In the combined set (n = 200), 67% BPNET/NEN (n = 134) had normal CgA levels. CgA levels did not distinguish histological subtypes (TC vs. AC, AUC 0.56 ± 0.04, p = 0.21), grade (p = 0.45-0.72), or progressive from stable disease (AUC 0.53 ± 0.05 p = 0.47). There was no correlation of CgA with Ki-67 index (Pearson r = 0.143, p = 0.14). For nonneoplastic diseases (chronic obstructive pulmonary disorder and idiopathic pulmonary fibrosis), CgA was elevated in 26-37%. For neoplastic disease (NSCLC, squamous cell carcinoma), CgA was elevated in 11-16%. The neuroendocrine SCLC also exhibited elevated CgA (50%). Elevated CgA was not useful for differentiating BPNET/NEN from these other pathologies. Monitoring BPNET/NEN over a 12-month period identified neither CgA levels per se nor changes in CgA were reflective of somatostatin analog treatment outcome/efficacy or the natural history of the disease (progression). CONCLUSIONS: Blood CgA levels are not clinically useful as a biomarker for lung BPNET/NEN. The low specificity and elevations in both nonneoplastic as well as other common neoplastic lung diseases identified limited clinical utility for this biomarker.

PRRT neuroendocrine tumor response monitored using circulating transcript analysis: the NETest.

PURPOSE: Peptide receptor radionuclide therapy (PRRT) is effective for metastatic/inoperable neuroendocrine tumors (NETs). Imaging response assessment is usually efficient subsequent to treatment completion. Blood biomarkers such as PRRT Predictive Quotient (PPQ) and NETest are effective in real-time. PPQ predicts PRRT efficacy; NETest monitors disease. We prospectively evaluated: (1) NETest as a surrogate biomarker for RECIST; (2) the correlation of NETest levels with PPQ prediction. METHODS: Three independent 177Lu-PRRT-treated GEP-NET and lung cohorts (Meldola, Italy: n = 72; Bad-Berka, Germany: n = 44; Rotterdam, Netherlands: n = 41). Treatment response: RECIST1.1 (responder (stable, partial, and complete response) vs non-responder). Blood sampling: pre-PRRT, before each cycle and follow-up (2-12 months). PPQ (positive/negative) and NETest (0-100 score) by PCR. Stable < 40; progressive > 40). CgA (ELISA) as comparator. Samples de-identified, measurement and analyses blinded. Kaplan-Meier survival and standard statistics. RESULTS: One hundred twenty-two of the 157 were evaluable. RECIST stabilization or response in 67%; 33% progressed. NETest significantly (p < 0.0001) decreased in RECIST "responders" (- 47 ± 3%); in "non-responders," it remained increased (+ 79 ± 19%) (p < 0.0005). NETest monitoring accuracy was 98% (119/122). Follow-up levels > 40 (progressive) vs stable (< 40) significantly correlated with mPFS (not reached vs. 10 months; HR 0.04 (95%CI, 0.02-0.07). PPQ response prediction was accurate in 118 (97%) with a 99% accurate positive and 93% accurate negative prediction. NETest significantly (p < 0.0001) decreased in PPQ-predicted responders (- 46 ± 3%) and remained elevated or increased in PPQ-predicted non-responders (+ 75 ± 19%). Follow-up NETest categories stable vs progressive significantly correlated with PPQ prediction and mPFS (not reached vs. 10 months; HR 0.06 (95%CI, 0.03-0.12). CgA did not reflect PRRT treatment: in RECIST responders decrease in 38% and in non-responders 56% (p = NS). CONCLUSIONS: PPQ predicts PRRT response in 97%. NETest accurately monitors PRRT response and is an effective surrogate marker of PRRT radiological response. NETest decrease identified responders and correlated (> 97%) with the pretreatment PPQ response predictor. CgA was non-informative.

The clinical applications of a multigene liquid biopsy (NETest) in neuroendocrine tumors.

PURPOSE: There are few effective biomarkers for neuroendocrine tumors. Precision oncology strategies have provided liquid biopsies for real-time and tailored decision-making. This has led to the development of the first neuroendocrine tumor liquid biopsy (the NETest). The NETest represents a transcriptomic signature of neuroendocrine tumor (NETs) that captures tumor biology and disease activity. The data have direct clinical application in terms of identifying residual disease, disease progress and the efficacy of treatment. In this overview we assess the available published information on the metrics and clinical efficacy of the NETest. MATERIAL AND METHODS: Published data on the NETest have been collated and analyzed to understand the clinical application of this multianalyte biomarker in NETs. RESULTS: NETest assay has been validated as a standardized and reproducible clinical laboratory measurement. It is not affected by demographic characteristics, or acid suppressive medication. Clinical utility of the NETest has been documented in gastroenteropancreatic, bronchopulmonary NETs, in paragangliomas and pheochromocytomas. The test facilitates accurate diagnosis of a NET disease, and real-time monitoring of the disease status (stable/progressive disease). It predicts aggressive tumor behavior, identifies operative tumor resection, and efficacy of the medical treatment (e.g. somatostatin analogues), or peptide receptor radionuclide therapy (PRRT). NETest metrics and clinical applications out-perform standard biomarkers like chromogranin A. CONCLUSIONS: The NETest exhibits clinically competent metrics as an effective biomarker for neuroendocrine tumors. Measurement of NET transcripts in blood is a significant advance in neuroendocrine tumor management and demonstrates that blood provides a viable source to identify and monitor tumor status.

Aryl Hydrocarbon Receptor Contributes to the Transcriptional Program of IL-10-Producing Regulatory B Cells.

Regulatory B cells (Bregs) play a critical role in the control of autoimmunity and inflammation. IL-10 production is the hallmark for the identification of Bregs. However, the molecular determinants that regulate the transcription of IL-10 and control the Breg developmental program remain unknown. Here, we demonstrate that aryl hydrocarbon receptor (AhR) regulates the differentiation and function of IL-10-producing CD19+CD21hiCD24hiBregs and limits their differentiation into B cells that contribute to inflammation. Chromatin profiling and transcriptome analyses show that loss of AhR in B cells reduces expression of IL-10 by skewing the differentiation of CD19+CD21hiCD24hiB cells into a pro-inflammatory program, under Breg-inducing conditions. B cell AhR-deficient mice develop exacerbated arthritis, show significant reductions in IL-10-producing Bregs and regulatory T cells, and show an increase in T helper (Th) 1 and Th17 cells compared with B cell AhR-sufficient mice. Thus, we identify AhR as a relevant contributor to the transcriptional regulation of Breg differentiation.

Utility of a ready-to-use PCR system for neuroendocrine tumor diagnosis.

BACKGROUND: Multigene-based PCR tests are time-consuming and limiting aspects of the protocol include increased risk of operator-based variation. In addition, such protocols are complex to transfer and reproduce between laboratories. AIMS: Evaluate the clinical utility of a pre-spotted PCR plate (PSP) for a novel multigene (n = 51) blood-based gene expression diagnostic assay for neuroendocrine tumors (NETs). METHODS: A pilot study (n = 44; 8 controls and 36 NETs) was undertaken to compare CQ, normalized gene expression and algorithm-based output (NETest score). Gene expression was then evaluated between matched blood:tumor tissue samples (n = 7). Thereafter, two prospective sets (diagnostic: n = 167; clinical validation: n = 48, respectively) were evaluated for diagnostic and clinical utility value. Two independent molecular diagnostics facilities were used to assess assay reproducibility and inter-laboratory metrics. Samples were collected (per CLIA protocol) processed to mRNA and cDNA and then either run per standard assay (liquid primers) or on PSPs. Separately, matching plasma samples were analyzed for chromogranin A (CgA). Statistics included non-parametric testing, Pearson-concordance, Predictive Modeling and AUROC analyses. RESULTS: In the pilot study (n = 44), CQ values were highly concordant (r: 0.82, p<0.0001) and normalized gene expression data significantly related (p<0.0001) (Pearson-pairwise correlation). NETest values were not different (49.7±33 standard vs. 48.5±31.5 PSP) and the overall concordance in output 96%. Predictive modelling confirmed this concordance (F1 score = 0.95). Gene expression levels were highly correlated between blood and tumor tissue (R: 0.71-0.83). In the diagnostic cohort (n = 30 controls, n = 87 non-NET controls, n = 50 NET), NETest was significantly lower (p<0.0001) in controls (11±6.5) and non-NET controls (13±18) than NETs (61±31). The AUROCs were 0.93-0.97 and the diagnostic accuracy was 90-97.5%. As a diagnostic, the PSP-NETest was significantly better than CgA (accuracy: 56%, p<0.0001). For clinical samples, the PSP generated robust and accurate (>96%) scores and was significantly better (p<0.0001) than CgA. The assay protocol was consistent (r: 0.97) and reproducible (co-efficient of variation: 1.3-4.2%) across the two facilities. CONCLUSION: The PSP protocol for the NETest has been established and prospectively tested in clinical samples. It is highly reproducible, has similar metrics (CV, categorization by control or NET) to the standard PCR assay and generates clinically concordant (>96%) NETest results. Moreover, it functions significantly more accurately than CgA.

Structure-preserving visualisation of high dimensional single-cell datasets.

Single-cell technologies offer an unprecedented opportunity to effectively characterize cellular heterogeneity in health and disease. Nevertheless, visualisation and interpretation of these multi-dimensional datasets remains a challenge. We present a novel framework, ivis, for dimensionality reduction of single-cell expression data. ivis utilizes a siamese neural network architecture that is trained using a novel triplet loss function. Results on simulated and real datasets demonstrate that ivis preserves global data structures in a low-dimensional space, adds new data points to existing embeddings using a parametric mapping function, and scales linearly to hundreds of thousands of cells. ivis is made publicly available through Python and R interfaces on https://github.com/beringresearch/ivis .

Measurement of circulating transcript levels (NETest) to detect disease recurrence and improve follow-up after curative surgical resection of well-differentiated pancreatic neuroendocrine tumors.

BACKGROUND: Recurrence of pancreatic neuroendocrine tumors (pNET) after surgery is common. Strategies to detect recurrence have limitations. We investigated the role of clinical criteria and the multigene polymerase chain reaction-based NETest during post-operative follow-up of pNET. METHODS: We studied 3 groups of resections: R0 with no recurrence (n = 11), R0 with recurrence (n = 12), and R1 with no recurrence (n = 12). NETest levels (>40%) were compared with chromogranin A (CgA) and clinicopathological criteria (CC; grade, lymph node metastases, size). Nonparametric, receiver operating characteristics, logistic regression, and predictive feature importance analyses were performed. RESULTS: NETest was higher in R0 with recurrence (56 ± 8%) compared with R1 with no recurrence (39 ± 6%) and R0 with no recurrence (28 ± 6%, P < .005). NETest positively correlated with recurrence (area under the curve: 0.82), CgA was not (area under the curve: 0.51 ± 0.09). Multiple regression analysis defined factor impact as highest for NETest (P < .005) versus CC (P < .03) and CgA (P = .23). NETest gave false positive or negative recurrence in 18% using a 40% cutoff. Logistic regression modeling of CC was 83% accurate; it was 91% when the NETest was included. Combining CC and NETest was approximately 2× more effective than individual CC alone (increase in R 2 value from 43% to 80%). CONCLUSIONS: A multigene blood test facilitates effective identification of pNET recurrence, prediction of disease relapse, and outperforms CgA.

The NETest: The Clinical Utility of Multigene Blood Analysis in the Diagnosis and Management of Neuroendocrine Tumors.

The neuroendocrine neoplasms test (NETest) is a multianalyte liquid biopsy that measures neuroendocrine tumor gene expression in blood. This unique signature precisely defines the biological activity of an individual tumor in real time. The assay meets the 3 critical requirements of an optimal biomarker: diagnostic accuracy, prognostic value, and predictive therapeutic assessment. NETest performance metrics are sensitivity and specificity and in head-to-head comparison are 4-fold to 10-fold more accurate than chromogranin A. NETest accurately identifies completeness of surgery and response to somatostatin analogs. Clinical registry data demonstrate significant clinical utility in watch/wait programs.

A liquid biopsy for bronchopulmonary/lung carcinoid diagnosis.

No effective blood biomarker exists to detect and clinically manage bronchopulmonary (BP) neuroendocrine tumors (NET). We developed a blood-based 51 NET-specific transcript set for diagnosis and monitoring and evaluated clinical performance metrics. It accurately diagnosed the tumor and differentiated stable from progressive disease as determined by RECIST criteria. Gene expression was evaluated in: a) publicly available BPNET transcriptomes (GSE35679); b) two BPNET cell-lines; and c) BPNET tissue with paired blood (n = 7). Blood gene expression was assessed in 194 samples including controls, benign lung diseases, malignant lung diseases and small bowel NETs. A separate validation study in 25 age- and gender-matched BPNETs/controls was performed. Gene expression measured by real-time PCR was scored (0-100%; normal: < 14%). Regression analyses, Principal Component Analysis (PCA), hierarchical clustering, Fisher's and non-parametric evaluations were undertaken. All 51 genes were identified in BPNET transcriptomes, tumor samples and cell-lines. Significant correlations were evident between paired tumor and blood (R2:0.63-0.91, p < 0.001). PCA and hierarchical clustering identified blood gene expression was significantly different between lung cancers and benign diseases, including BPNETs. Gene expression was highly correlated (R2: 0.91, p = 1.7 × 10-15) between small bowel and BPNET. For validation, all 25 BPNETs were positive compared to 20% controls (p < 0.0001). Scores were significantly elevated (p < 0.0001) in BPNETs (57 ± 28%) compared to controls (4 ± 5%). BPNETs with progressive disease (85 ± 11%) exhibited higher scores than stable disease (32 ± 7%, p < 0.0001). Blood measurements accurately diagnosed bronchopulmonary carcinoids, distinguishing stable from progressive disease. This marker panel will have clinical utility as a diagnostic liquid biopsy able to define disease activity and progression in real-time.

PRRT genomic signature in blood for prediction of 177Lu-octreotate efficacy.

BACKGROUND: Peptide receptor radionuclide therapy (PRRT) utilizes somatostatin receptor (SSR) overexpression on neuroendocrine tumors (NET) to deliver targeted radiotherapy. Intensity of uptake at imaging is considered related to efficacy but has low sensitivity. A pretreatment strategy to determine individual PRRT response remains a key unmet need. NET transcript expression in blood integrated with tumor grade provides a PRRT predictive quotient (PPQ) which stratifies PRRT "responders" from "non-responders". This study clinically validates the utility of the PPQ in NETs. METHODS: The development and validation of the PPQ was undertaken in three independent 177Lu-PRRT treated cohorts. Specificity was tested in two separate somatostatin analog-treated cohorts. Prognostic value of the marker was defined in a cohort of untreated patients. The developmental cohort included lung and gastroenteropancreatic [GEP] NETs (n = 72) from IRST Meldola, Italy. The majority were GEP (71%) and low grade (86% G1-G2). Prospective validation cohorts were from Zentralklinik Bad Berka, Germany (n = 44), and Erasmus Medical Center, Rotterdam, Netherlands (n = 42). Each cohort included predominantly well differentiated, low grade (86-95%) lung and GEP-NETs. The non-PRRT comparator cohorts included SSA cohort I, n = 28 (100% low grade, 100% GEP-NET); SSA cohort II, n = 51 (98% low grade; 76% GEP-NET); and an untreated cohort, n = 44 (64% low grade; 91% GEP-NET). Baseline evaluations included clinical information (disease status, grade, SSR) and biomarker (CgA). NET blood gene transcripts (n = 8: growth factor signaling and metabolism) were measured pre-therapy and integrated with tumor Ki67 using a logistic regression model. This provided a binary output: "predicted responder" (PPQ+); "predicted non-responder" (PPQ-). Treatment response was evaluated using RECIST criteria [Responder (stable, partial and complete response) vs Non-Responder)]. Sample measurement and analyses were blinded to study outcome. Statistical evaluation included Kaplan-Meier survival and standard test evaluation analyses. RESULTS: In the developmental cohort, 56% responded to PRRT. The PPQ predicted 100% of responders and 84% of non-responders (accuracy: 93%). In the two validation cohorts (response: 64-79%), the PPQ was 95% accurate (Bad Berka: PPQ + =97%, PPQ- = 93%; Rotterdam: PPQ + =94%, PPQ- = 100%). Overall, the median PFS was not reached in PPQ+ vs PPQ- (10-14 months; HR: 18-77, p < 0.0001). In the comparator cohorts, the predictor (PPQ) was 47-50% accurate for SSA-treatment and 50% as a prognostic. No differences in PFS were respectively noted (PPQ+: 10-12 months vs. PPQ-: 9-15 months). CONCLUSION: The PPQ derived from circulating NET specific genes and tumor grade prior to the initiation of therapy is a highly specific predictor of the efficacy of PRRT with an accuracy of 95%.