Using machine learning to improve anaphylaxis case identification in medical claims data.

Objectives: Anaphylaxis is a severe life-threatening allergic reaction, and its accurate identification in healthcare databases can harness the potential of "Big Data" for healthcare or public health purposes. Materials and methods: This study used claims data obtained between October 1, 2015 and February 28, 2019 from the CMS database to examine the utility of machine learning in identifying incident anaphylaxis cases. We created a feature selection pipeline to identify critical features between different datasets. Then a variety of unsupervised and supervised methods were used (eg, Sammon mapping and eXtreme Gradient Boosting) to train models on datasets of differing data quality, which reflects the varying availability and potential rarity of ground truth data in medical databases. Results: Resulting machine learning model accuracies ranged from 47.7% to 94.4% when tested on ground truth data. Finally, we found new features to help experts enhance existing case-finding algorithms. Discussion: Developing precise algorithms to detect medical outcomes in claims can be a laborious and expensive process, particularly for conditions presented and coded diversely. We found it beneficial to filter out highly potent codes used for data curation to identify underlying patterns and features. To improve rule-based algorithms where necessary, researchers could use model explainers to determine noteworthy features, which could then be shared with experts and included in the algorithm. Conclusion: Our work suggests machine learning models can perform at similar levels as a previously published expert case-finding algorithm, while also having the potential to improve performance or streamline algorithm construction processes by identifying new relevant features for algorithm construction.

Using machine learning to improve anaphylaxis case identification in medical claims data.

Objective: Anaphylaxis is a severe life-threatening allergic reaction, and its accurate identification in healthcare databases can harness the potential of "Big Data" for healthcare or public health purposes. Methods: This study used claims data obtained between October 1, 2015 and February 28, 2019 from the CMS database to examine the utility of machine learning in identifying incident anaphylaxis cases. We created a feature selection pipeline to identify critical features between different datasets. Then a variety of unsupervised and supervised methods were used (eg, Sammon mapping and eXtreme Gradient Boosting) to train models on datasets of differing data quality, which reflects the varying availability and potential rarity of ground truth data in medical databases. Results: Resulting machine learning model accuracies ranged between 47.7% and 94.4% when tested on ground truth data. Finally, we found new features to help experts enhance existing case-finding algorithms. Discussion: Developing precise algorithms to detect medical outcomes in claims can be a laborious and expensive process, particularly for conditions presented and coded diversely. We found it beneficial to filter out highly potent codes used for data curation to identify underlying patterns and features. To improve rule-based algorithms where necessary, researchers could use model explainers to determine noteworthy features, which could then be shared with experts and included in the algorithm. Conclusion: Our work suggests machine learning models can perform at similar levels as a previously published expert case-finding algorithm, while also having the potential to improve performance or streamline algorithm construction processes by identifying new relevant features for algorithm construction.

Increased PPARD Expression May Play a Protective Role in Human Lung Adenocarcinoma and Squamous Cell Carcinoma.

Peroxisome proliferator-activated receptor-δ, encoded by gene PPARD, is overexpressed in a majority of human lung cancer subtypes, but its role in the tumor progression remains poorly understood. We have analyzed the expression of PPARD in lung adenocarcinoma (LA) and squamous cell carcinoma (LSCC) datasets. The potential roles of PPARD in the pathological development of LA and LSCC were explored through literature-based pathway analysis and pathway enrichment analysis. In all LA datasets (N = 11) and in seven out of nine LSCC studies, the levels of PPARD were increased as compared to control tissues (log-fold changes were 0.37 ± 0.20 and 0.10 ± 0.37 for LA and LSCC, respectively). On average, the expression levels of PPARD in LA were higher than those in LSCC (p = 0.036). Pathway analysis showed that the overexpression of PPARD might play both positive and negative roles in the development of both LA and LSCC. Specifically, PPARD inhibits seven LSCC promoters and seven LA promoters and activates one LSCC inhibitor and another LA inhibitor. However, PPARD also activates six and one promoters of LA and LSCC, respectively, which would facilitate the development of LA/LSCC. Our results suggested a mixed role of PPARD in LA/LSCC, which may add new insights into the understanding of the PPARD-lung cancer relationship.

Schizophrenia Plays a Negative Role in the Pathological Development of Myocardial Infarction at Multiple Biological Levels.

It has shown that schizophrenia (SCZ) is associated with a higher chance of myocardial infarction (MI) and increased mortality. However, the underlying mechanism is largely unknown. Here, we first constructed a literature-based genetic pathway linking SCZ and MI, and then we tested the expression levels of the genes involved in the pathway by a meta-analysis using nine gene expression datasets of MI. In addition, a literature-based data mining process was conducted to explore the connection between SCZ at different levels: small molecules, complex molecules, and functional classes. The genetic pathway revealed nine genes connecting SCZ and MI. Specifically, SCZ activates two promoters of MI (IL6 and CRP) and deactivates seven inhibitors of MI (ADIPOQ, SOD2, TXN, NGF, ADORA1, NOS1, and CTNNB1), suggesting that no protective role of SCZ in MI was detected. Meta-analysis showed that one promoter of MI (CRP) presented no significant increase, and six out of seven genetic inhibitors of MI demonstrated minor to moderately increased expression. Therefore, the elevation of CRP and inhibition of the six inhibitors of MI by SCZ could be critical pathways to promote MI. Nine other regulators of MI were influenced by SCZ, including two gene families (inflammatory cytokine and IL1 family), five small molecules (lipid peroxide, superoxide, ATP, ascorbic acid, melatonin, arachidonic acid), and two complexes (CaM kinase 2 and IL23). Our results suggested that SCZ promotes the development and progression of MI at different levels, including genes, small molecules, complex molecules, and functional classes.

Explore the Underlying Mechanism Between Atopic Dermatitis and Major Depressive Disorder.

Adult patients with atopic dermatitis (AD) present relatively higher rates of major depressive disorder (MDD). However, the underlying mechanism is largely unknown. Here, we first conducted a systematic literature-based data mining to identify entities linking AD and MDD, including proteins, cells, functional classes, and small molecules. Then we conducted an AD-RNA expression data-based mega-analysis to test the expression variance of the genes that were regulators of MDD. After that, a Fisher Exact test-based pathway enrichment analysis (PEA) was performed to explore the AD-driven MDD-genetic regulators' functionality. We identified 22 AD-driven entities that were up-stream MDD regulators, including 11 genes, seven small molecules, three functional classes, and one cell. AD could exert a promoting effect on the development of MDD. Four of the 11 genes demonstrated significant expression changes in AD patients in favor of the development of MDD. PEA results showed that AD mainly drives cytokine/chemokine regulation and neuroinflammatory response-related pathways to influence the pathological development of MDD. Our results supported the promotion role of AD in the pathological development of MDD, including the regulation of multiple genetic regulators of MDD involved in cytokine/chemokine regulation and inflammatory response.

PPARD May Play a Protective Role for Major Depressive Disorder.

Activation of PPARD has been shown to inhibit depressive behaviors and enhances neurogenesis. However, whether PPARD is involved in the pathological development of major depressive disorder (MDD) is largely unknown. To explore the potential connection between PPARD and MDD, we first conducted a literature-based data mining to construct a PPARD-driven MDD regulating network. Then, we tested the PPARD expression changes in MDD patients from 18 independent MDD RNA expression datasets, followed by coexpression analysis, multiple linear regression analysis, and a heterogeneity analysis to study the influential factors for PPARD expression levels. Our results showed that overexpression of PPARD could inhibit inflammatory cytokine signaling pathways and the ROS and glutamate pathways that have been shown to play important roles in the pathological development of MDD. However, PPARD could also activate nitric oxide formation and ceramide synthesis, which was implicated as promoters in the pathogenesis of MDD, indicating the complexity of the relationship between PPARD and MDD. PPARG presented significant within- and between-study variations in the 18 MDD datasets (p value = 0.97), which were significantly associated with the population region (country) and sample source (p < 2.67e - 5). Our results suggested that PPARD could be a potential regulator rather than a biomarker in the pathological development of MDD. This study may add new insights into the understanding of the PPARD-MDD relationship.

Integrating Literature-Based Knowledge Database and Expression Data to Explore Molecular Pathways Connecting PPARG and Myocardial Infarction.

Peroxisome proliferator-activated receptor γ (PPARG) might play a protective role in the development of myocardial infarction (MI) with limited mechanisms identified. Genes associated with both PPARG and MI were extracted from Elsevier Pathway Studio to construct the initial network. The gene expression activity within the network was estimated through a mega-analysis with eight independent expression datasets derived from Gene Expression Omnibus (GEO) to build PPARG and MI connecting pathways. After that, gene set enrichment analysis (GSEA) was conducted to explore the functional profile of the genes involved in the PPARG-driven network. PPARG demonstrated a significantly low expression in MI patients (LFC = -0.52; p < 1.84e - 9). Consequently, PPARG could indicatively be promoting three MI inhibitors (e.g., SOD1, CAV1, and POU5F1) and three MI-downregulated markers (e.g., ALB, ACADM, and ADIPOR2), which were deactivated in MI cases (p < 0.05), and inhibit two MI-upregulated markers (RELA and MYD88), which showed increased expression levels in MI cases (p = 0.0077 and 0.047, respectively). These eight genes were mainly enriched in nutrient- and cell metabolic-related pathways and functionally linked by GSEA and PPCN. Our results suggest that PPARG could protect the heart against both the development and progress of MI through the regulation of nutrient- and metabolic-related pathways.

Aging-associated genes TNFRSF12A and CHI3L1 contribute to thyroid cancer: An evidence for the involvement of hypoxia as a driver.

The prevalence of thyroid cancer (TC) is high in the elderly. The present study was based on the hypothesis that genes, which have increased activity with aging, may play a role in the development of TC. A large-scale literature-based data analysis was conducted to explore the genes that are implicated in both TC and aging. Subsequently, a mega-analysis of 16 RNA expression datasets (1,222 samples: 439 healthy controls, and 783 patients with TC) was conducted to test a set of genes associated with aging but not TC. To uncover a possible link between these genes and TC, a functional pathway analysis was conducted, and the results were validated by analysis of gene co-expression. A multiple linear regression (MLR) model was employed to study the possible influence of sample size, population region and study age on the gene expression levels in TC. A total of 262 and 816 genes were identified to have increased activity with aging and TC, respectively; with a significant overlap of 63 genes (P<3.82×10-35). The mega-analysis revealed two aging-associated genes (CHI3L1 and TNFRSF12A) to be significantly associated with TC (P<2.05×10-8), and identified the association with multiple hypoxia-driven pathways through functional pathway analysis, also confirmed by the co-expression analysis. The MLR analysis identified population region as a significant factor contributing to the expression levels of CHI3L1 and TNFRSF12A in TC samples (P<3.24×10-4). The determination of genes that promote aging was warranted due to their possible involvement in TC. The present study suggests CHI3L1 and TNFRSF12A as novel common risk genes associated with both aging and TC.

Integrative analysis of shared genetic pathogenesis by autism spectrum disorder and obsessive-compulsive disorder.

Many common pathological features have been observed for both autism spectrum disorders (ASDs) and obsessive-compulsive disorder (OCD). However, no systematic analysis of the common gene markers associated with both ASD and OCD has been conducted so far. Here, two batches of large-scale literature-based disease-gene relation data (updated in 2017 and 2019, respectively) and gene expression data were integrated to study the possible association between OCD and ASD at the genetic level. Genes linked to OCD and ASD present significant overlap (P-value <2.64e-39). A genetic network of over 20 genes was constructed, through which OCD and ASD may exert influence on each other. The 2017-based analysis suggested six potential common risk genes for OCD and ASD (CDH2, ADCY8, APOE, TSPO, TOR1A, and OLIG2), and the 2019-based study identified two more genes (DISP1 and SETD1A). Notably, the gene APOE identified by the 2017-based analysis has been implicated to have an association with ASD in a recent study (2018) with DNA methylation analysis. Our results support the possible complex genetic associations between OCD and ASD. Genes linked to one disease are worth further investigation as potential risk factors for the other.

Pathways of aging: comparative analysis of gene signatures in replicative senescence and stress induced premature senescence.

BACKGROUND: In culturing normal diploid cells, senescence may either happen naturally, in the form of replicative senescence, or it may be a consequence of external challenges such as oxidative stress. Here we present a comparative analysis aimed at reconstruction of molecular cascades specific for replicative (RS) and stressinduced senescence (SIPS) in human fibroblasts. RESULTS: An involvement of caspase-3/keratin-18 pathway and serine/threonine kinase Aurora A/ MDM2 pathway was shared between RS and SIPS. Moreover, stromelysin/MMP3 and N-acetylglucosaminyltransferase enzyme MGAT1, which initiates the synthesis of hybrid and complex Nglycans, were identified as key orchestrating components in RS and SIPS, respectively. In RS only, Aurora-B driven cell cycle signaling was deregulated in concert with the suppression of anabolic branches of the fatty acids and estrogen metabolism. In SIPS, Aurora-B signaling is deprioritized, and the synthetic branches of cholesterol metabolism are upregulated, rather than downregulated. Moreover, in SIPS, proteasome/ubiquitin ligase pathways of protein degradation dominate the regulatory landscape. This picture indicates that SIPS proceeds in cells that are actively fighting stress which facilitates premature senescence while failing to completely activate the orderly program of RS. The promoters of genes differentially expressed in either RS or SIPS are unusually enriched by the binding sites for homeobox family proteins, with particular emphasis on HMX1, IRX2, HDX and HOXC13. Additionally, we identified Iroquois Homeobox 2 (IRX2) as a master regulator for the secretion of SPP1-encoded osteopontin, a stromal driver for tumor growth that is overexpressed by both RS and SIPS fibroblasts. The latter supports the hypothesis that senescence-specific de-repression of SPP1 aids in SIPS-dependent stromal activation. CONCLUSIONS: Reanalysis of previously published experimental data is cost-effective approach for extraction of additional insignts into the functioning of biological systems.