Towards accelerating model parallelism in distributed deep learning systems.

Modern deep neural networks cannot be often trained on a single GPU due to large model size and large data size. Model parallelism splits a model for multiple GPUs, but making it scalable and seamless is challenging due to different information sharing among GPUs with communication overhead. Specifically, we identify two key issues to make the parallelism being inefficient and inaccurate; an efficient pipelining technique is crucial to maximize GPU utilization and normalizations in deep neural networks may affect the performance due to different statistics sharing of mini-batch. In this work, we address these issues by investigating efficient pipelining for model parallelism and effective normalizations in model / data parallelisms when training a model with large mini-batch in multiple GPUs so that the model performance in accuracy can not be compromised. Firstly, we propose a novel method to search for an optimal micro-batch size considering the number of GPUs and memory size for model parallelism. For efficient pipelining, mini-batch is usually divided into smaller batches (called micro-batch). To maximize the utilization of GPU computing resources, training should be performed with the optimal micro-batch size. Our proposed micro-batch size search algorithm achieved increased image throughput by up to 12% and improved trainable mini-batch size by 25% as compared to the conventional model parallelism method. Secondly, we investigate normalizations in distributed deep learning training for different parallelisms. Our experiments using different normalization methods suggested that the performance with batch normalization can be improved by sharing the batch information among GPUs when performing data parallelism. It was also confirmed that group normalization helped minimizing accuracy degradation when performing model parallelism with pipelining and yielded consistent accuracies for diverse mini-batch sizes.

Cytomegalovirus reactivation under pre-emptive therapy following allogeneic hematopoietic stem cell transplant: Pattern, survival, and risk factors in the Republic of Korea.

INTRODUCTION: Pre-emptive therapy for cytomegalovirus (CMV) reactivation has been used in allogeneic hematopoietic stem cell transplantation (allo-HSCT). It is unclear if this strategy has poorer clinical outcomes in CMV-endemic areas and if more aggressive prophylaxis is required. METHODS: We retrospectively analyzed the patterns and survival after CMV reactivation in patients undergoing pre-emptive therapy following allo-HSCT and assessed high-risk patients who could benefit from aggressive CMV prophylaxis in endemic areas. RESULTS: Of the 292 patients who underwent allo-HSCT, 70.5% (donor+ or recipient+) were CMV seropositive. CMV reactivation occurred in 139 patients (47.6%), with a median of 31.5 days from day 0 of allo-HSCT. The overall survival of patients with CMV reactivation who received pre-emptive therapy did not differ from those without reactivation. Of the 139 patients with CMV reactivation, 78 (56.1%) underwent ≥2 rounds of pre-emptive therapy. In multivariate analysis, the risk of CMV reactivation was higher in patients with multiple myeloma, with CMV seropositivity of the recipient and donor, administered with a higher dose of anti-thymocyte globulin (ATG), and with acute graft-versus-host disease (aGVHD) ≥ grade 2. CONCLUSION: Although half of the patients with allo-HSCT were administered with pre-emptive therapy for CMV, CMV reactivation did not affect their survival, indicating the advantages of pre-emptive therapy, even in CMV-endemic areas. The cost-effectiveness of more aggressive CMV prophylaxis should be re-evaluated in patients at a high risk for CMV reactivation.

Prognostic outcomes of cytomegalovirus reactivation after autologous stem cell transplantation.

Background: Cytomegalovirus (CMV) reactivation is a common complication in patients undergoing allogeneic stem cell transplantation. However, the incidence of CMV reactivation is low after autologous stem cell transplantation (auto-SCT), and the prognostic value of CMV reactivation remains controversial. Moreover, reports on late CMV reactivation after auto-SCT are limited. We aimed to analyze the association between CMV reactivation and survival outcomes and develop a predictive model for late CMV reactivation in patients undergoing auto-SCT. Methods: Data of 201 patients who underwent SCT at the Korea University Medical Center from 2007 to 2018 were collected. We analyzed prognostic factors for survival outcomes after auto-SCT and risk factors for late CMV reactivation using a receiver operating characteristic curve. Then, we developed a predictive risk model for late CMV reactivation based on results of the risk factor analysis. Results: Early CMV reactivation was significantly associated with better overall survival (OS) (hazard ratio [HR], 0.329; P = 0.045) in patients with multiple myeloma; however, no significant differences were observed in patients with lymphoma. For late CMV reactivation, a serum lactate dehydrogenase level greater than the upper limit of normal (HR, 2.251; P = 0.027) and late CMV reactivation (HR, 2.964; P = 0.047) were independent risk factors for poor OS, while lymphoma diagnosis (vs. multiple myeloma; HR, 0.389; P = 0.016) was an independent risk factor for good OS. In risk factor analysis for late CMV reactivation, T-cell lymphoma diagnosis (odds ratio [OR], 8.499; P = 0.029), ≥ two prior chemotherapies (OR, 8.995; P = 0.027), failure to achieve complete response (CR) after transplantation (OR, 7.124; P = 0.031), and early CMV reactivation (OR, 12.853; P = 0.007) were significantly associated with late CMV reactivation. To develop the predictive risk model for late CMV reactivation, a score (1 to 1.5) was assigned for each of the above-mentioned variables. The optimal cutoff value (1.75 points) was calculated using the receiver operating characteristic curve. The predictive risk model showed good discrimination, with an area under the curve of 0.872 (standard error, 0.062; P < 0.001). Conclusions: Late CMV reactivation was an independent risk factor for inferior OS, whereas early CMV reactivation was associated with better survival in patients with multiple myeloma. This risk prediction model could be helpful in identifying high-risk patients who require monitoring for late CMV reactivation and potentially benefit from prophylactic or preemptive therapy.

Improving Voluntary Compliance through Problem-Oriented Governance: A Case Study of South Korea's Efforts to Increase the COVID-19 Vaccination Rate.

Drawing on the literature of problem-oriented governance (POG) and social motivation for voluntary compliance, this study discusses how South Korea's efforts to cultivate distributed cognitions and build core capabilities of POG-reflective improvement, collaborative, and data analytic capabilities-contribute to the citizens' voluntary compliance with the current vaccination policy by improving trust and confidence. A systematic content analysis and documentation review of relevant policies, situation reports, after-action reports, official briefings, and news articles provide significant implications for both theories and practices of policy compliance and governance for effective and efficient management of many wicked problems like the COVID-19 pandemic.

Blood Transfusion, All-Cause Mortality and Hospitalization Period in COVID-19 Patients: Machine Learning Analysis of National Health Insurance Claims Data.

This study presents the most comprehensive machine-learning analysis for the predictors of blood transfusion, all-cause mortality, and hospitalization period in COVID-19 patients. Data came from Korea National Health Insurance claims data with 7943 COVID-19 patients diagnosed during November 2019−May 2020. The dependent variables were all-cause mortality and the hospitalization period, and their 28 independent variables were considered. Random forest variable importance (GINI) was introduced for identifying the main factors of the dependent variables and evaluating their associations with these predictors, including blood transfusion. Based on the results of this study, blood transfusion had a positive association with all-cause mortality. The proportions of red blood cell, platelet, fresh frozen plasma, and cryoprecipitate transfusions were significantly higher in those with death than in those without death (p-values < 0.01). Likewise, the top ten factors of all-cause mortality based on random forest variable importance were the Charlson Comorbidity Index (53.54), age (45.68), socioeconomic status (45.65), red blood cell transfusion (27.08), dementia (19.27), antiplatelet (16.81), gender (14.60), diabetes mellitus (13.00), liver disease (11.19) and platelet transfusion (10.11). The top ten predictors of the hospitalization period were the Charlson Comorbidity Index, socioeconomic status, dementia, age, gender, hemiplegia, antiplatelet, diabetes mellitus, liver disease, and cardiovascular disease. In conclusion, comorbidity, red blood cell transfusion, and platelet transfusion were the major factors of all-cause mortality based on machine learning analysis. The effective management of these predictors is needed in COVID-19 patients.

Generation of the human pluripotent stem cell lines KUMi005-A from a patients with multiple myeloma.

Multiple myeloma (MM) progresses with abnormal monoclonal proliferation and accumulation of malignant plasma cells in the bone marrow. We established human induced pluripotent stem cells (iPSCs), KUMi005-A, from bone marrow samples of a patient with MM. This reprogrammed cell line has similar characteristics to human embryonic stem cells, such as proliferation properties and pluripotency. KUMi005-A iPSCs may be applicable in MM disease modeling and cell-based therapies.

Generation of human induced iPSC KUMi003-A from acute promyelocytic leukemia (APL) M3.

Acute promyelocytic leukemia (APL) M3 is an acute myeloid leukemia (AML) subtype and is characterized by the chromosomal translocation t(15;17)(p22;q11), which results in the fusion of the promyelocytic gene (PML) at 15q22 with the retinoic acid α-receptor gene (RARA) at 17q21. We generated an induced pluripotent stem cell line "KUMi003-A" from an APL M3 patient that is pluripotent and can differentiate into the three germ layers. This iPSC line will be useful as a disease model to investigate disease mechanisms specific to APL M3.

PVR (CD155) Expression as a Potential Prognostic Marker in Multiple Myeloma.

Poliovirus receptor (PVR, CD155) is upregulated during tumor progression, and PVR expression is associated with poor prognosis in cancer patients; however, prognostic implications for PVR in multiple myeloma (MM) have not been investigated. PVR plays an immunomodulatory role by interacting with CD226, CD96, and TIGIT. TIGIT is a checkpoint inhibitory receptor that can limit adaptive and innate immunity, and it binds to PVR with the highest affinity. We used immunohistochemistry, ELISA, qPCR, and flow cytometry to investigate the role of PVR in MM. PVR was highly expressed in patients with MM, and membrane PVR expression showed a significant correlation with soluble PVR levels. PVR expression was significantly associated with the Revised-International Staging System stage, presence of extramedullary plasmacytoma and bone lesion, percentage of bone marrow plasma cells (BMPCs), and ß2-microglobulin levels, suggesting a possible role in advanced stages and metastasis. Furthermore, TIGIT expression was significantly correlated with the percentage of BMPCs. Patients with high PVR expression had significantly shorter overall and progression-free survival, and PVR expression was identified as an independent prognostic factor for poor MM survival. These findings indicate that PVR expression is associated with MM stage and poor prognosis, and is a potential prognostic marker for MM.

Generation of a human induced pluripotent stem cell line KUMi006 from a patient with multiple myeloma.

In this study, we report the generation of a novel human induced pluripotent stem cell (hiPSC) line from bone marrow mononuclear cells of a patient with multiple myeloma, using an integrative Sendai virus vector. This pluripotent cell line has been shown to differentiate into three germ layers. Therefore, these induced pluripotent stem cells (iPSCs) will enable not only advances in cell therapy products but also the study of mechanisms.

Postoperative Thromboembolism According to the Type of Surgery: A Nationwide Study in the Republic of Korea.

Postoperative thromboembolism (TE) is a serious, but preventable, complication in surgical patients. Orthopedic surgery, neurosurgery, and vascular surgery are considered high risk for TE, and current guidelines recommend TE prophylaxis. However, insufficient data exist regarding TE risk in other general surgeries. This study identified the actual incidence and relative risk of postoperative TE in the real world, according to surgery type. Twenty-six surgeries between 1 December 2017 and 31 August 2019 were selected from the Health Insurance Review and Assessment Service database and analyzed for postoperative TE events. Among all patients, 2.17% had a TE event within 6 months of surgery and 0.75% had a TE event owing to anticoagulant treatment. The incidence of total TE events was the highest in total knee replacement (12.77%), hip replacement (11.46%), and spine surgery (5.98%). The incidence of TE with anticoagulant treatment was the highest in total knee replacement (7.40%), hip replacement (7.20%), and coronary artery bypass graft (CABG) surgery (3.81%). Hip replacement, total knee replacement, CABG surgery, spine surgery, and cardiac surgery except CABG surgery, showed relatively higher risks for total claimed venous TE. The relative risk of venous TE with anticoagulant treatment was the highest for hysterectomy, partial hepatectomy, hip replacement, cardiac surgery except CABG surgery, and total knee replacement. The relative risk of arterial TE was the highest for cardiac surgery, total knee replacement, and hip replacement. In the real world, the incidence of postoperative TE events from total knee replacement and those from hip replacement remain high, and some surgeries could have a relatively higher risk of TE than other surgeries. For patients undergoing these surgeries, studies to reduce the incidence of postoperative TE in clinical practice should be conducted.

A human pluripotent stem cell line KUMi004-A generated from a patient with chronic lymphocytic leukemia.

Chronic lymphocytic leukemia (CLL) is a type of blood cancer caused by the abnormal accumulation of malignant plasma cells. In this study, we generated CLL iPSCs (KUMi004-a) using the Sendai virus, confirming pluripotency. Also, it can differentiate into three primary germ layers. We expect this cell line could be helpful to understand the pathology of CLL.

Clinical Efficacy of Ruxolitinib in Patients with Myelofibrosis: A Nationwide Population-Based Study in Korea.

Previous studies have reported the survival benefit after ruxolitinib treatment in patients with myelofibrosis (MF). However, population-based data of its efficacy are limited. We analyzed the effects of ruxolitinib in MF patients with data from the Korean National Health Insurance Database. In total, 1199 patients diagnosed with MF from January 2011 to December 2017 were identified, of which 731 were included in this study. Patients who received ruxolitinib (n = 224) were matched with those who did not receive the drug (n = 507) using the 1:1 greedy algorithm. Propensity scores were formulated using five variables: age, sex, previous history of arterial/venous thrombosis, and red blood cell (RBC) or platelet (PLT) transfusion dependence at the time of diagnosis. Cox regression analysis for overall survival (OS) revealed that ruxolitinib treatment (hazard ratio (HR), 0.67; p = 0.017) was significantly related to superior survival. In the multivariable analysis for OS, older age (HR, 1.07; p < 0.001), male sex (HR, 1.94; p = 0.021), and RBC (HR, 3.72; p < 0.001) or PLT (HR, 9.58; p = 0.001) transfusion dependence were significantly associated with poor survival, although type of MF did not significantly affect survival. Considering evidence supporting these results remains weak, further studies on the efficacy of ruxolitinib in other populations are needed.

The effect of the response to the coronavirus disease pandemic on treatment outcomes in patients with lymphoma and multiple myeloma.

BACKGROUND/AIMS: Relatively little data are available on how the response to the coronavirus disease 2019 (COVID-19) pandemic has affected treatment outcomes in patients receiving chemotherapy for lymphoma or multiple myeloma. We aimed to determine the effect of COVID-19 countermeasures on treatment outcomes in this patient population. METHODS: We retrospectively analyzed data on patients treated for lymphoma or multiple myeloma in two tertiary hospitals in Seoul. Patients were divided into two groups: group 1 included patients who received chemotherapy between September and December 2019 (the control period), and group 2 included patients who received chemotherapy between September and December 2020 (the study period). Countermeasures to COVID-19 were applied to the patients in group 2. The countermeasures implemented included mask wearing and regular handwashing at home and in hospital; COVID-19 risk assessments on all hospital visitors; and pre-emptive COVID-19 screening for all newly hospitalized patients and their resident guardians. RESULTS: No differences in treatment outcomes, including treatment response, incidence and duration of neutropenia or neutropenic fever, delays in chemotherapy, or number of deaths during chemotherapy, were observed between the g roups. None of the patients in group 2 tested positive for COVID-19, and there were no COVID-19-related deaths during the study period. CONCLUSION: Countermeasures to COVID-19 did not affect treatment outcomes in patients receiving chemotherapy for lymphoma or multiple myeloma. Data on the effect of countermeasures to COVID-19 on treatment outcomes should continue to be analyzed to ensure that treatment outcomes are not adversely affected.

A nationwide study of patients with monoclonal gammopathy of undetermined significance with a 10-year follow-up in South Korea.

In clinical practice, most patients with monoclonal gammopathy of undetermined significance (MGUS) undergo long-term follow-up without disease progression. There is insufficient real-world data about how closely and whether anything other than disease progression should be monitored. Herein, we performed a nationwide study of 470 patients with MGUS with a 10-year follow-up to determine the patterns of disease progression and other comorbidities. During the follow-up period, 158 of 470 patients with MGUS (33.62%) progressed to symptomatic monoclonal gammopathies. Most of these were multiple myeloma (134/470 patients, 28.51%), and those diagnosed within 2 years after diagnosis of MGUS was high. Approximately 30-50% of patients with MGUS had hypertension, diabetes, hyperlipidemia, and osteoarthritis at the time of diagnosis, and these comorbidities were newly developed during the follow-up period in approximately 50% of the remaining patients with MGUS. Approximately 20-40% of patients with MGUS have acute or chronic kidney failure, thyroid disorders, disc disorders, peripheral neuropathy, myocardial infarction, stroke, and heart failure during the follow-up period. Altogether, when MGUS is diagnosed, close follow-up of the possibility of progression to multiple myeloma is required, especially within 2 years after diagnosis; simultaneously, various comorbidities should be considered and monitored during the follow-up of patients with MGUS. Continuous research is needed to establish appropriate follow-up guidelines.

Prediction Model for Cereblon Expression in Bone Marrow Plasma Cells Based on Blood Markers in Multiple Myeloma Patients.

BACKGROUND: Cereblon (CRBN) is a direct target of immunomodulatory drugs (IMiDs) and is known to be sensitive and responsive to IMiD therapy. We evaluated CRBN expression in bone marrow plasma cells and analyzed whether CRBN expression was associated with multiple myeloma prognosis. Lastly, we developed a nomogram model for predicting high CRBN expression based on clinically significant blood markers. METHODS: We evaluated 143 multiple myeloma patients (internal dataset) who underwent bone marrow examinations. For evaluating the prognostic ability of the nomogram model, two external cohorts (235 patients in external dataset 1 and 156 patients in external dataset 2) were analyzed. The expression of CRBN in bone marrow aspirate samples was evaluated using immunohistochemistry. High CRBN expression was defined as the study-defined H-score ≥6. RESULTS: In the high CRBN group, the median progression-free survival (PFS) and overall survival (OS) of patients receiving the IMiD-based therapy and non-IMiD therapy were 29 and 10 months for PFS, and NR (not reached) and 54 months for OS, respectively. IMiD-based therapy was significantly associated with better PFS and OS outcomes. High CRBN expression was independently predicted by female sex, high serum free-light chain (FLC) ratio, higher serum M-protein level, and higher ß2-microglobulin level. Based on these results, we constructed a new nomogram model to predict high CRBN expression and the effectiveness of IMiD therapy in multiple myeloma. CONCLUSION: This nomogram could improve the prognostic evaluation of myeloma patients exhibiting high CRBN expression treated with IMiD therapy and might help provide personalized treatment strategies to clinicians.

Prognostic Value of Galectin-9 Relates to Programmed Death-Ligand 1 in Patients With Multiple Myeloma.

Galectin-9 (Gal-9) expression can be negatively or positively associated with cancer patient prognosis, depending on the cancer type. However, the nature of this relationship remains unclear in multiple myeloma. Therefore, we evaluated the prognostic value of Gal-9 and its relationship with the expression of PD-L1 molecule, the most widely studied immune checkpoint inhibitor, in patients with newly diagnosed multiple myeloma. Gal-9 and PD-L1 levels in bone marrow aspirate samples were evaluated using immunofluorescence assays. Gal-9 positivity was defined as having ≥1% Gal-9-expressing plasma cells. PD-L1 expression was categorized as low or high based on its median value. The median OS of patients with positive and negative Gal-9 expression was 42 months and not reached, respectively. However, no significant difference was observed in OS between the two groups (P = 0.10). Patients with high PD-L1 expression had OS times of 14 and 43 months in the positive and negative Gal-9 expression groups, respectively. In the high PD-L1 expression group, patients expressing Gal-9 had significantly worse OS than those negative for it (P = 0.019). Multivariable Cox analysis confirmed that Gal-9 expression could independently predict shortened OS (hazard ratio, 1.090; 95% confidence interval, 1.015-1.171; P = 0.018) in patients with high PD-L1 expression. However, in the low PD-L1 expression group, patients with high Gal-9 expression exhibited a trend toward better OS (P = 0.816). Our results indicate that the prognostic value of Gal-9 may be related to PD-L1 expression in patients with newly diagnosed multiple myeloma.

Generation of induced pluripotent stem cell line KUMi002-A with normal karyotype from a patient with Philadelphia chromosome-positive chronic myeloid leukemia.

Chronic myeloid leukemia (CML) is caused by the dysregulated tyrosine kinase activity of the BCR-ABL fusion protein. In this study, we generated induced pluripotent stem cells (iPSCs) with a normal karyotype, using cells from a patient with CML and a Philadelphia chromosome. These human iPSCs showed positive pluripotency markers and differentiated into three germ layers. This iPSC line can be useful for the study of CML, namely the biology of hematopoietic stem cells with normal karyotype in CML, and for the development of patient-specific immunological treatment.

Generation of the induced pluripotent stem cell line KUMi001-A carrying the Philadelphia chromosome from a chronic myeloid leukemia patient.

Chronic myeloid leukemia (CML) is caused by the BCR-ABL fusion protein, which dysregulates tyrosine kinase activity. In this study, we generated induced pluripotent stem cells (iPSCs) carrying the Philadelphia chromosome from a CML patient with the BCR-ABL fusion protein. CML iPSCs were positive for pluripotency markers and had the ability to differentiate into the three germ layers. This iPSC cell line could be useful for studying CML pathogenesis as well as for drug development to treat CML.

The Role of Nitric Oxide-Induced ATILL6 in Growth and Disease Resistance in Arabidopsis thaliana.

Nitric oxide (NO) is a signaling molecule that regulates various processes, including plant growth and development, immunity, and environmental interactions. Using high throughput RNA-seq data, we explored the role of the NO-induced ATILL6 gene in plant growth and defense using functional genomics. The atill6 mutant and wild-types were challenged with either oxidative (H2O2, MV) or nitro-oxidative (CySNO, GSNO) stress conditions, and the phenotypic results showed that ATILL6 gene differentially regulates cotyledon development frequency (CDF) as well as the root and shoot lengths of the plants. To investigate whether ATILL6 plays a role in plant basal or resistance (R)-gene-mediated defense, the plants were challenged with either virulent or avirulent strains of Pseudomonas syringae pathovar tomato (Pst) DC3000. The atill6 line showed a susceptible phenotype, higher pathogen growth, and highly reduced transcript accumulation of PR1 and PR2 genes. These results suggested that ATILL6 positively regulates plant basal defense. Furthermore, after the inoculation of atill6 with avirulent Pst (DC3000), the expressions of the PR1 and PR2 genes decreased, suggesting a positive role in R-gene-mediated resistance in protecting the plant from further spread of disease. We also investigated the role of ATILL6 in systemic acquired resistance (SAR), and the results showed that ATILL6 positively regulates SAR, as the mutant line atill6 has significantly (p ≤ 0.05) lower transcript accumulation of PR, G3DPH, and AZI genes. Overall, these results indicate that the NO-induced ATILL6 gene differentially regulates plant growth and positively regulates plant basal defense, R-gene-mediated resistance, and SAR.