Automated Deep Learning-Based Diagnosis and Molecular Characterization of Acute Myeloid Leukemia Using Flow Cytometry.

The current flow cytometric analysis of blood and bone marrow samples for diagnosis of acute myeloid leukemia (AML) relies heavily on manual intervention in the processing and analysis steps, introducing significant subjectivity into resulting diagnoses and necessitating highly trained personnel. Furthermore, concurrent molecular characterization via cytogenetics and targeted sequencing can take multiple days, delaying patient diagnosis and treatment. Attention-based multi-instance learning models (ABMILMs) are deep learning models that make accurate predictions and generate interpretable insights regarding the classification of a sample from individual events/cells; nonetheless, these models have yet to be applied to flow cytometry data. In this study, we developed a computational pipeline using ABMILMs for the automated diagnosis of AML cases based exclusively on flow cytometric data. Analysis of 1820 flow cytometry samples shows that this pipeline provides accurate diagnoses of acute leukemia (area under the receiver operating characteristic curve [AUROC] 0.961) and accurately differentiates AML vs B- and T-lymphoblastic leukemia (AUROC 0.965). Models for prediction of 9 cytogenetic aberrancies and 32 pathogenic variants in AML provide accurate predictions, particularly for t(15;17)(PML::RARA) [AUROC 0.929], t(8;21)(RUNX1::RUNX1T1) (AUROC 0.814), and NPM1 variants (AUROC 0.807). Finally, we demonstrate how these models generate interpretable insights into which individual flow cytometric events and markers deliver optimal diagnostic utility, providing hematopathologists with a data visualization tool for improved data interpretation, as well as novel biological associations between flow cytometric marker expression and cytogenetic/molecular variants in AML. Our study is the first to illustrate the feasibility of using deep learning-based analysis of flow cytometric data for automated AML diagnosis and molecular characterization.

An Automated Pipeline for Differential Cell Counts on Whole-Slide Bone Marrow Aspirate Smears.

The pathologic diagnosis of bone marrow disorders relies in part on the microscopic analysis of bone marrow aspirate (BMA) smears and the manual counting of marrow nucleated cells to obtain a differential cell count (DCC). This manual process has significant limitations, including the analysis of only a small subset of optimal slide areas and nucleated cells, as well as interobserver variability due to differences in cell selection and classification. To address these shortcomings, we developed an automated machine learning-based pipeline for obtaining 11-component DCCs on whole-slide BMAs. This pipeline uses a sequential process of identifying optimal BMA regions with high proportions of marrow nucleated cells, detecting individual cells within these optimal areas, and classifying these cells into 1 of 11 DCC components. Convolutional neural network models were trained on 396,048 BMA region, 28,914 cell boundary, and 1,510,976 cell class images from manual annotations. The resulting automated pipeline produced 11-component DCCs that demonstrated a high statistical and diagnostic concordance with manual DCCs among a heterogeneous group of testing BMA slides with varying pathologies and cellularities. Additionally, we demonstrated that an automated analysis can reduce the intraslide variance in DCCs by analyzing the whole slide and marrow nucleated cells within all optimal regions. Finally, the pipeline outputs of region classification, cell detection, and cell classification can be visualized using whole-slide image analysis software. This study demonstrates the feasibility of a fully automated pipeline for generating DCCs on scanned whole-slide BMA images, with the potential for improving the current standard of practice for utilizing BMA smears in the laboratory analysis of hematologic disorders.

The Immunophenotypic Profile of Healthy Human Bone Marrow.

Flow cytometry enables multiparametric characterization of hematopoietic cell immunophenotype. Deviations from normal immunophenotypic patterns comprise a cardinal feature of many hematopoietic neoplasms, underscoring the ongoing essentiality of flow cytometry as a diagnostic tool. However, understanding of aberrant hematopoiesis requires an equal understanding of normal hematopoiesis as a comparator. In this review, we outline key features of healthy adult hematopoiesis and lineage specification as illuminated by flow cytometry and provide diagrams illustrating what a diagnostician may observe in flow cytometric plots. These features provide a profile of baseline hematopoiesis, to which clinical samples with suspected neoplasia may be compared.

Digital assessment of peripheral blood and bone marrow aspirate smears.

The diagnosis of benign and neoplastic hematologic disorders relies on analysis of peripheral blood and bone marrow aspirate smears. As demonstrated by the widespread laboratory adoption of hematology analyzers for automated assessment of peripheral blood, digital analysis of these samples provides many significant benefits compared to relying solely on manual review. Nonetheless, analogous instruments for digital bone marrow aspirate smear assessment have yet to be clinically implemented. In this review, we first provide a historical overview detailing the implementation of hematology analyzers for digital peripheral blood assessment in the clinical laboratory, including the improvements in accuracy, scope, and throughput of current instruments over prior generations. We also describe recent research in digital peripheral blood assessment, particularly in the development of advanced machine learning models that may soon be incorporated into commercial instruments. Next, we provide an overview of recent research in digital assessment of bone marrow aspirate smears and how these approaches could soon lead to development and clinical adoption of instrumentation for automated bone marrow aspirate smear analysis. Finally, we describe the relative advantages and provide our vision for the future of digital assessment of peripheral blood and bone marrow aspirate smears, including what improvements we can soon expect in the hematology laboratory.

Automated Deep Learning-Based Diagnosis and Molecular Characterization of Acute Myeloid Leukemia using Flow Cytometry.

Current flow cytometric analysis of blood and bone marrow samples for diagnosis of acute myeloid leukemia (AML) relies heavily on manual intervention in both the processing and analysis steps, introducing significant subjectivity into resulting diagnoses and necessitating highly trained personnel. Furthermore, concurrent molecular characterization via cytogenetics and targeted sequencing can take multiple days, delaying patient diagnosis and treatment. Attention-based multi-instance learning models (ABMILMs) are deep learning models which make accurate predictions and generate interpretable insights regarding the classification of a sample from individual events/cells; nonetheless, these models have yet to be applied to flow cytometry data. In this study, we developed a computational pipeline using ABMILMs for the automated diagnosis of AML cases based exclusively on flow cytometric data. Analysis of 1,820 flow cytometry samples shows that this pipeline provides accurate diagnoses of acute leukemia [AUROC 0.961] and accurately differentiates AML versus B- and T-lymphoblastic leukemia [AUROC 0.965]. Models for prediction of 9 cytogenetic aberrancies and 32 pathogenic variants in AML provide accurate predictions, particularly for t(15;17)(PML::RARA) [AUROC 0.929], t(8;21)(RUNX1::RUNX1T1) [AUROC 0.814], and NPM1 variants [AUROC 0.807]. Finally, we demonstrate how these models generate interpretable insights into which individual flow cytometric events and markers deliver optimal diagnostic utility, providing hematopathologists with a data visualization tool for improved data interpretation, as well as novel biological associations between flow cytometric marker expression and cytogenetic/molecular variants in AML. Our study is the first to illustrate the feasibility of using deep learning-based analysis of flow cytometric data for automated AML diagnosis and molecular characterization.

Lessons learned: A look back at the performance of nine COVID-19 serologic assays and their proposed utility.

BACKGROUND: Serologic assays for the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) have been proposed to assist with the acute diagnosis of infection, support epidemiological studies, identify convalescent plasma donors, and evaluate vaccine response. METHODS: We report an evaluation of nine serologic assays: Abbott (AB) and Epitope (EP) IgG and IgM, EUROIMMUN (EU) IgG and IgA, Roche anti-N (RN TOT) and anti-S (RS TOT) total antibody, and DiaSorin (DS) IgG. We evaluated 291 negative controls (NEG CTRL), 91 PCR positive (PCR POS) patients (179 samples), 126 convalescent plasma donors (CPD), 27 healthy vaccinated donors (VD), and 20 allogeneic hematopoietic stem cell transplant (HSCT) recipients (45 samples). RESULTS: We observed good agreement with the method performance claims for specificity (93-100%) in NEG CTRL but only 85% for EU IgA. The sensitivity claims in the first 2 weeks of symptom onset was lower (26-61%) than performance claims based on > 2 weeks since PCR positivity. We observed high sensitivities (94-100%) in CPD except for AB IgM (77%), EP IgM (0%). Significantly higher RS TOT was observed for Moderna vaccine recipients then Pfizer (p-values < 0.0001). A sustained RS TOT response was observed for the five months following vaccination. HSCT recipients demonstrated significantly lower RS TOT than healthy VD (p < 0.0001) at dose 2 and 4 weeks after. CONCLUSIONS: Our data suggests against the use of anti-SARS-CoV-2 assays to aid in acute diagnosis. RN TOT and RS TOT can readily identify past-resolved infection and vaccine response in the absence of native infection. We provide an estimate of expected antibody response in healthy VD over the time course of vaccination for which to compare antibody responses in immunosuppressed patients.

Integration of machine learning and genome-scale metabolic modeling identifies multi-omics biomarkers for radiation resistance.

Resistance to ionizing radiation, a first-line therapy for many cancers, is a major clinical challenge. Personalized prediction of tumor radiosensitivity is not currently implemented clinically due to insufficient accuracy of existing machine learning classifiers. Despite the acknowledged role of tumor metabolism in radiation response, metabolomics data is rarely collected in large multi-omics initiatives such as The Cancer Genome Atlas (TCGA) and consequently omitted from algorithm development. In this study, we circumvent the paucity of personalized metabolomics information by characterizing 915 TCGA patient tumors with genome-scale metabolic Flux Balance Analysis models generated from transcriptomic and genomic datasets. Metabolic biomarkers differentiating radiation-sensitive and -resistant tumors are predicted and experimentally validated, enabling integration of metabolic features with other multi-omics datasets into ensemble-based machine learning classifiers for radiation response. These multi-omics classifiers show improved classification accuracy, identify clinical patient subgroups, and demonstrate the utility of personalized blood-based metabolic biomarkers for radiation sensitivity. The integration of machine learning with genome-scale metabolic modeling represents a significant methodological advancement for identifying prognostic metabolite biomarkers and predicting radiosensitivity for individual patients.

Personalized Genome-Scale Metabolic Models Identify Targets of Redox Metabolism in Radiation-Resistant Tumors.

Redox cofactor production is integral toward antioxidant generation, clearance of reactive oxygen species, and overall tumor response to ionizing radiation treatment. To identify systems-level alterations in redox metabolism that confer resistance to radiation therapy, we developed a bioinformatics pipeline for integrating multi-omics data into personalized genome-scale flux balance analysis models of 716 radiation-sensitive and 199 radiation-resistant tumors. These models collectively predicted that radiation-resistant tumors reroute metabolic flux to increase mitochondrial NADPH stores and reactive oxygen species (ROS) scavenging. Simulated genome-wide knockout screens agreed with experimental siRNA gene knockdowns in matched radiation-sensitive and radiation-resistant cancer cell lines, revealing gene targets involved in mitochondrial NADPH production, central carbon metabolism, and folate metabolism that allow for selective inhibition of glutathione production and H2O2 clearance in radiation-resistant cancers. This systems approach represents a significant advancement in developing quantitative genome-scale models of redox metabolism and identifying personalized metabolic targets for improving radiation sensitivity in individual cancer patients.

MTHFD2 Blockade Enhances the Efficacy of ß-Lapachone Chemotherapy With Ionizing Radiation in Head and Neck Squamous Cell Cancer.

Head and Neck Squamous Cell Cancer (HNSCC) presents with multiple treatment challenges limiting overall survival rates and affecting patients' quality of life. Amongst these, resistance to radiation therapy constitutes a major clinical problem in HNSCC patients compounded by origin, location, and tumor grade that limit tumor control. While cisplatin is considered the standard radiosensitizing agent for definitive or adjuvant radiotherapy, in recurrent tumors or for palliative care other chemotherapeutics such as the antifolates methotrexate or pemetrexed are also being utilized as radiosensitizers. These drugs inhibit the enzyme dihydrofolate reductase, which is essential for DNA synthesis and connects the 1-C/folate metabolism to NAD(P)H and NAD(P)+ balance in cells. In previous studies, we identified MTHFD2, a mitochondrial enzyme involved in folate metabolism, as a key contributor to NAD(P)H levels in the radiation-resistant cells and HNSCC tumors. In the study presented here, we investigated the role of MTHFD2 in the response to radiation alone and in combination with ß-lapachone, a NQO1 bioactivatable drug, which generates reactive oxygen species concomitant with NAD(P)H oxidation to NAD(P)+. These studies are performed in a matched HNSCC cell model of response to radiation: the radiation resistant rSCC-61 and radiation sensitive SCC-61 cells reported earlier by our group. Radiation resistant rSCC-61 cells had increased sensitivity to ß-lapachone compared to SCC-61 and knockdown of MTHFD2 in rSCC-61 cells further potentiated the cytotoxicity of ß-lapachone with radiation in a dose and time-dependent manner. rSCC-61 MTHFD2 knockdown cells irradiated and treated with ß-lapachone showed increased PARP1 activation, inhibition of mitochondrial respiration, decreased respiration-linked ATP production, and increased mitochondrial superoxide and protein oxidation as compared to control rSCC-61 scrambled shRNA. Thus, these studies point to MTHFD2 as a potential target for development of radiosensitizing chemotherapeutics and potentiator of ß-lapachone cytotoxicity.

An Integrative Gene Expression and Mathematical Flux Balance Analysis Identifies Targetable Redox Vulnerabilities in Melanoma Cells.

Melanomas harboring BRAF mutations can be treated with BRAF inhibitors (BRAFi), but responses are varied and tumor recurrence is inevitable. Here we used an integrative approach of experimentation and mathematical flux balance analyses in BRAF-mutated melanoma cells to discover that elevated antioxidant capacity is linked to BRAFi sensitivity in melanoma cells. High levels of antioxidant metabolites in cells with reduced BRAFi sensitivity confirmed this conclusion. By extending our analyses to other melanoma subtypes in The Cancer Genome Atlas, we predict that elevated redox capacity is a general feature of melanomas, not previously observed. We propose that redox vulnerabilities could be exploited for therapeutic benefits and identify unsuspected combination targets to enhance the effects of BRAFi in any melanoma, regardless of mutational status. SIGNIFICANCE: An integrative bioinformatics, flux balance analysis, and experimental approach identify targetable redox vulnerabilities and show the potential for modulation of cancer antioxidant defense to augment the benefits of existing therapies in melanoma.

Peroxiredoxins in Cancer and Response to Radiation Therapies.

Peroxiredoxins have a long-established cellular function as regulators of redox metabolism by catalyzing the reduction of peroxides (e.g., H2O2, lipid peroxides) with high catalytic efficiency. This activity is also critical to the initiation and relay of both phosphorylation and redox signaling in a broad range of pathophysiological contexts. Under normal physiological conditions, peroxiredoxins protect normal cells from oxidative damage that could promote oncogenesis (e.g., environmental stressors). In cancer, higher expression level of peroxiredoxins has been associated with both tumor growth and resistance to radiation therapies. However, this relationship between the expression of peroxiredoxins and the response to radiation is not evident from an analysis of data in The Cancer Genome Atlas (TCGA) or NCI60 panel of cancer cell lines. The focus of this review is to summarize the current experimental knowledge implicating this class of proteins in cancer, and to provide a perspective on the value of targeting peroxiredoxins in the management of cancer. Potential biases in the analysis of the TCGA data with respect to radiation resistance are also highlighted.

Targeting NAD+ Metabolism to Enhance Radiation Therapy Responses.

Nicotinamide adenine dinucleotide (NAD+) metabolism is integrally connected with the mechanisms of action of radiation therapy and is altered in many radiation-resistant tumors. This makes NAD+ metabolism an ideal target for therapies that increase radiation sensitivity and improve patient outcomes. This review provides an overview of NAD+ metabolism in the context of the cellular response to ionizing radiation, as well as current therapies that target NAD+ metabolism to enhance radiation therapy responses. Additionally, we summarize state-of-the-art methods for measuring, modeling, and manipulating NAD+ metabolism, which are being used to identify novel targets in the NAD+ metabolic network for therapeutic interventions in combination with radiation therapy.

Genome-Scale Modeling of NADPH-Driven ß-Lapachone Sensitization in Head and Neck Squamous Cell Carcinoma.

AIMS: The purpose of this study was to investigate differential nicotinamide adenine dinucleotide phosphate, reduced (NADPH) production between radiation-sensitive and -resistant head and neck squamous cell carcinoma (HNSCC) cell lines and whether these differences are predictive of sensitivity to the chemotherapeutic ß-lapachone. RESULTS: We have developed a novel human genome-scale metabolic modeling platform that combines transcriptomic, kinetic, thermodynamic, and metabolite concentration data. Upon incorporation of this information into cell line-specific models, we observed that the radiation-resistant HNSCC model redistributed flux through several major NADPH-producing reactions. Upon RNA interference of canonical NADPH-producing genes, the metabolic network can further reroute flux through alternate NADPH biosynthesis pathways in a cell line-specific manner. Model predictions of perturbations in cellular NADPH production after gene knockdown match well with experimentally verified effects of ß-lapachone treatment on NADPH/NADP+ ratio and cell viability. This computational approach accurately predicts HNSCC-specific oxidoreductase genes that differentially affect cell viability between radiation-responsive and radiation-resistant cancer cells upon ß-lapachone treatment. INNOVATION: Quantitative genome-scale metabolic models that incorporate multiple levels of biological data are applied to provide accurate predictions of responses to a NADPH-dependent redox cycling chemotherapeutic drug under a variety of perturbations. CONCLUSION: Our modeling approach suggests differences in metabolism and ß-lapachone redox cycling that underlie phenotypic differences in radiation-sensitive and -resistant cancer cells. This approach can be extended to investigate the synergistic action of NAD(P)H: quinone oxidoreductase 1 bioactivatable drugs and radiation therapy. Antioxid. Redox Signal. 29, 937-952.

Predicting clinical outcomes from large scale cancer genomic profiles with deep survival models.

Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general high-dimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an open-source software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.