Open-source machine learning pipeline automatically flags instances of acute respiratory distress syndrome from electronic health records.

Physicians could greatly benefit from automated diagnosis and prognosis tools to help address information overload and decision fatigue. Intensive care physicians stand to benefit greatly from such tools as they are at particularly high risk for those factors. Acute Respiratory Distress Syndrome (ARDS) is a life-threatening condition affecting >10% of critical care patients and has a mortality rate over 40%. However, recognition rates for ARDS have been shown to be low (30-70%) in clinical settings. In this work, we present a reproducible computational pipeline that automatically adjudicates ARDS on retrospective datasets of mechanically ventilated adult patients. This pipeline automates the steps outlined by the Berlin Definition through implementation of natural language processing tools and classification algorithms. We train an XGBoost model on chest imaging reports to detect bilateral infiltrates, and another on a subset of attending physician notes labeled for the most common ARDS risk factor in our data. Both models achieve high performance-a minimum area under the receiver operating characteristic curve (AUROC) of 0.86 for adjudicating chest imaging reports in out-of-bag test sets, and an out-of-bag AUROC of 0.85 for detecting a diagnosis of pneumonia. We validate the entire pipeline on a cohort of MIMIC-III encounters and find a sensitivity of 93.5% - an extraordinary improvement over the 22.6% ARDS recognition rate reported for these encounters - along with a specificity of 73.9%. We conclude that our reproducible, automated diagnostic pipeline exhibits promising accuracy, generalizability, and probability calibration, thus providing a valuable resource for physicians aiming to enhance ARDS diagnosis and treatment strategies. We surmise that proper implementation of the pipeline has the potential to aid clinical practice by facilitating the recognition of ARDS cases at scale.

Meta-Research: Understudied genes are lost in a leaky pipeline between genome-wide assays and reporting of results.

Present-day publications on human genes primarily feature genes that already appeared in many publications prior to completion of the Human Genome Project in 2003. These patterns persist despite the subsequent adoption of high-throughput technologies, which routinely identify novel genes associated with biological processes and disease. Although several hypotheses for bias in the selection of genes as research targets have been proposed, their explanatory powers have not yet been compared. Our analysis suggests that understudied genes are systematically abandoned in favor of better-studied genes between the completion of -omics experiments and the reporting of results. Understudied genes remain abandoned by studies that cite these -omics experiments. Conversely, we find that publications on understudied genes may even accrue a greater number of citations. Among 45 biological and experimental factors previously proposed to affect which genes are being studied, we find that 33 are significantly associated with the choice of hit genes presented in titles and abstracts of -omics studies. To promote the investigation of understudied genes, we condense our insights into a tool, find my understudied genes (FMUG), that allows scientists to engage with potential bias during the selection of hits. We demonstrate the utility of FMUG through the identification of genes that remain understudied in vertebrate aging. FMUG is developed in Flutter and is available for download at fmug.amaral.northwestern.edu as a MacOS/Windows app.

Modern techniques for studying human genetics have helped to identify 20,000 protein-encoding genes in the human genome. Yet scientists have not studied most of them, including genes linked to human diseases in genome wide studies. For example, about 44% of the genes associated with Alzheimer's disease have never been mentioned in the title or summary of a scientific article. Why so many health-linked genes have yet to be examined is unclear. Many genetic studies instead focus on genes already studied before the Human Genome Project mapped the entire genome in 2003. There are many reasons why scientists may ignore potentially disease-causing genes. They may feel that well-studied genes are safer bets or more likely to result in high-profile publications. Or they may lack the tools to study less well-characterized genes. Richardson et al. analyzed the scientific literature for clues on why so many genes are being ignored by scientists. The analysis included hundreds of articles that used a wide range of genetic techniques, including genome-wide association studies, RNA sequencing, and gene editing tools to scour the genome for disease-linked genes. It revealed that scientists abandon the study of many genes early in the research process and identify 33 reasons why. Contrary to scientists' fears, Richardson et al. show that reports on understudied genes often garner more attention than studies on well-known genes. Richardson et al. used their results to create a downloadable tool called "Find My Understudied Genes (FMUG)" to help scientists identify understudied genes and counteract bias toward more well-studied genes. The app may help scientists make informed decisions about which understudied genes to research. If the tool helps boost investigation of understudied genes, it may help speed up progress towards understanding human genetics and how various genes may contribute to diseases.

Meta-Research: understudied genes are lost in a leaky pipeline between genome-wide assays and reporting of results.

Present-day publications on human genes primarily feature genes that already appeared in many publications prior to completion of the Human Genome Project in 2003. These patterns persist despite the subsequent adoption of high-throughput technologies, which routinely identify novel genes associated with biological processes and disease. Although several hypotheses for bias in the selection of genes as research targets have been proposed, their explanatory powers have not yet been compared. Our analysis suggests that understudied genes are systematically abandoned in favor of better-studied genes between the completion of -omics experiments and the reporting of results. Understudied genes remain abandoned by studies that cite these -omics experiments. Conversely, we find that publications on understudied genes may even accrue a greater number of citations. Among 45 biological and experimental factors previously proposed to affect which genes are being studied, we find that 33 are significantly associated with the choice of hit genes presented in titles and abstracts of - omics studies. To promote the investigation of understudied genes we condense our insights into a tool, find my understudied genes (FMUG), that allows scientists to engage with potential bias during the selection of hits. We demonstrate the utility of FMUG through the identification of genes that remain understudied in vertebrate aging. FMUG is developed in Flutter and is available for download at fmug.amaral.northwestern.edu as a MacOS/Windows app.

Energy metabolism modulates the regulatory impact of activators on gene expression.

Gene expression is a regulated process fueled by ATP consumption. Therefore, regulation must be coupled to constraints imposed by the level of energy metabolism. Here, we explore this relationship both theoretically and experimentally. A stylized mathematical model predicts that activators of gene expression have variable impact depending on metabolic rate. Activators become less essential when metabolic rate is reduced and more essential when metabolic rate is enhanced. We find that, in the Drosophila eye, expression dynamics of the yan gene are less affected by loss of EGFR-mediated activation when metabolism is reduced, and the opposite effect is seen when metabolism is enhanced. The effects are also seen at the level of pattern regularity in the adult eye, where loss of EGFR-mediated activation is mitigated by lower metabolism. We propose that gene activation is tuned by energy metabolism to allow for faithful expression dynamics in the face of variable metabolic conditions.

Energy metabolism modulates the regulatory impact of activators on gene expression.

Gene expression is a regulated process fueled by ATP consumption. Therefore, regulation must be coupled to constraints imposed by the level of energy metabolism. Here, we explore this relationship both theoretically and experimentally. A stylized mathematical model predicts that activators of gene expression have variable impact depending on metabolic rate. Activators become less essential when metabolic rate is reduced and more essential when metabolic rate is enhanced. We find that in the Drosophila eye, expression dynamics of the yan gene are less affected by loss of EGFR-mediated activation when metabolism is reduced, and the opposite effect is seen when metabolism is enhanced. The effects are also seen at the level of pattern regularity in the adult eye, where loss of EGFR-mediated activation is mitigated by lower metabolism. We propose that gene activation is tuned by energy metabolism to allow for faithful expression dynamics in the face of variable metabolic conditions.

Validation of a tool for estimating clinician recognition of ARDS using data from the international LUNG SAFE study.

Under-recognition of acute respiratory distress syndrome (ARDS) by clinicians is an important barrier to adoption of evidence-based practices such as low tidal volume ventilation. The burden created by the COVID-19 pandemic makes it even more critical to develop scalable data-driven tools to improve ARDS recognition. The objective of this study was to validate a tool for accurately estimating clinician ARDS recognition rates using discrete clinical characteristics easily available in electronic health records. We conducted a secondary analysis of 2,705 ARDS and 1,261 non-ARDS hypoxemic patients in the international LUNG SAFE cohort. The primary outcome was validation of a tool that estimates clinician ARDS recognition rates from health record data. Secondary outcomes included the relative impact of clinical characteristics on tidal volume delivery and clinician documentation of ARDS. In both ARDS and non-ARDS patients, greater height was associated with lower standardized tidal volume (mL/kg PBW) (ARDS: adjusted ß = -4.1, 95% CI -4.5 --3.6; non-ARDS: ß = -7.7, 95% CI -8.8 --6.7, P<0.00009 [where α = 0.01/111 with the Bonferroni correction]). Standardized tidal volume has already been normalized for patient height, and furthermore, height was not associated with clinician documentation of ARDS. Worsening hypoxemia was associated with both increased clinician documentation of ARDS (ß = -0.074, 95% CI -0.093 --0.056, P<0.00009) and lower standardized tidal volume (ß = 1.3, 95% CI 0.94-1.6, P<0.00009) in ARDS patients. Increasing chest imaging opacities, plateau pressure, and clinician documentation of ARDS also were associated with lower tidal volume in ARDS patients. Our EHR-based data-driven approach using height, gender, ARDS documentation, and lowest standardized tidal volume yielded estimates of clinician ARDS recognition rates of 54% for mild, 63% for moderate, and 73% for severe ARDS. Our tool replicated clinician-reported ARDS recognition in the LUNG SAFE study, enabling the identification of ARDS patients at high risk of being unrecognized. Our approach can be generalized to other conditions for which there is a need to increase adoption of evidence-based care.

Ratiometric sensing of Pnt and Yan transcription factor levels confers ultrasensitivity to photoreceptor fate transitions in Drosophila.

Cell state transitions are often triggered by large changes in the concentrations of transcription factors and therefore large differences in their stoichiometric ratios. Whether cells can elicit transitions using modest changes in the ratios of co-expressed factors is unclear. Here, we investigate how cells in the Drosophila eye resolve state transitions by quantifying the expression dynamics of the ETS transcription factors Pnt and Yan. Eye progenitor cells maintain a relatively constant ratio of Pnt/Yan protein, despite expressing both proteins with pulsatile dynamics. A rapid and sustained twofold increase in the Pnt/Yan ratio accompanies transitions to photoreceptor fates. Genetic perturbations that modestly disrupt the Pnt/Yan ratio produce fate transition defects consistent with the hypothesis that transitions are normally driven by a twofold shift in the ratio. A biophysical model based on cooperative Yan-DNA binding coupled with non-cooperative Pnt-DNA binding illustrates how twofold ratio changes could generate ultrasensitive changes in target gene transcription to drive fate transitions. Thus, coupling cell state transitions to the Pnt/Yan ratio sensitizes the system to modest fold-changes, conferring robustness and ultrasensitivity to the developmental program.

A new approach for extracting information from protein dynamics.

Increased ability to predict protein structures is moving research focus towards understanding protein dynamics. A promising approach is to represent protein dynamics through networks and take advantage of well-developed methods from network science. Most studies build protein dynamics networks from correlation measures, an approach that only works under very specific conditions, instead of the more robust inverse approach. Thus, we apply the inverse approach to the dynamics of protein dihedral angles, a system of internal coordinates, to avoid structural alignment. Using the well-characterized adhesion protein, FimH, we show that our method identifies networks that are physically interpretable, robust, and relevant to the allosteric pathway sites. We further use our approach to detect dynamical differences, despite structural similarity, for Siglec-8 in the immune system, and the SARS-CoV-2 spike protein. Our study demonstrates that using the inverse approach to extract a network from protein dynamics yields important biophysical insights.

Forecasting the evolution of fast-changing transportation networks using machine learning.

Transportation networks play a critical role in human mobility and the exchange of goods, but they are also the primary vehicles for the worldwide spread of infections, and account for a significant fraction of CO2 emissions. We investigate the edge removal dynamics of two mature but fast-changing transportation networks: the Brazilian domestic bus transportation network and the U.S. domestic air transportation network. We use machine learning approaches to predict edge removal on a monthly time scale and find that models trained on data for a given month predict edge removals for the same month with high accuracy. For the air transportation network, we also find that models trained for a given month are still accurate for other months even in the presence of external shocks. We take advantage of this approach to forecast the impact of a hypothetical dramatic reduction in the scale of the U.S. air transportation network as a result of policies to reduce CO2 emissions. Our forecasting approach could be helpful in building scenarios for planning future infrastructure.

The first step is recognizing there is a problem: a methodology for adjusting for variability in disease severity when estimating clinician performance.

BACKGROUND: Adoption of innovations in the field of medicine is frequently hindered by a failure to recognize the condition targeted by the innovation. This is particularly true in cases where recognition requires integration of patient information from different sources, or where disease presentation can be heterogeneous and the recognition step may be easier for some patients than for others. METHODS: We propose a general data-driven metric for clinician recognition that accounts for the variability in patient disease severity and for institutional standards. As a case study, we evaluate the ventilatory management of 362 patients with acute respiratory distress syndrome (ARDS) at a large academic hospital, because clinician recognition of ARDS has been identified as a major barrier to adoption to evidence-based ventilatory management. We calculate our metric for the 48 critical care physicians caring for these patients and examine the relationships between differences in ARDS recognition performance from overall institutional levels and provider characteristics such as demographics, social network position, and self-reported barriers and opinions. RESULTS: Our metric was found to be robust to patient characteristics previously demonstrated to affect ARDS recognition, such as disease severity and patient height. Training background was the only factor in this study that showed an association with physician recognition. Pulmonary and critical care medicine (PCCM) training was associated with higher recognition (ß = 0.63, 95% confidence interval 0.46-0.80, p < 7 × 10- 5). Non-PCCM physicians recognized ARDS cases less frequently and expressed greater satisfaction with the ability to get the information needed for making an ARDS diagnosis (p < 5 × 10- 4), suggesting that lower performing clinicians may be less aware of institutional barriers. CONCLUSIONS: We present a data-driven metric of clinician disease recognition that accounts for variability in patient disease severity and for institutional standards. Using this metric, we identify two unique physician populations with different intervention needs. One population consistently recognizes ARDS and reports barriers vs one does not and reports fewer barriers.

A new approach for extracting information from protein dynamics.

Increased ability to predict protein structures is moving research focus towards understanding protein dynamics. A promising approach is to represent protein dynamics through networks and take advantage of well-developed methods from network science. Most studies build protein dynamics networks from correlation measures, an approach that only works under very specific conditions, instead of the more robust inverse approach. Thus, we apply the inverse approach to the dynamics of protein dihedral angles, a system of internal coordinates, to avoid structural alignment. Using the well-characterized adhesion protein, FimH, we show that our method identifies networks that are physically interpretable, robust, and relevant to the allosteric pathway sites. We further use our approach to detect dynamical differences, despite structural similarity, for Siglec-8 in the immune system, and the SARS-CoV-2 spike protein. Our study demonstrates that using the inverse approach to extract a network from protein dynamics yields important biophysical insights.

The characteristics of early-stage research into human genes are substantially different from subsequent research.

Throughout the last 2 decades, several scholars observed that present day research into human genes rarely turns toward genes that had not already been extensively investigated in the past. Guided by hypotheses derived from studies of science and innovation, we present here a literature-wide data-driven meta-analysis to identify the specific scientific and organizational contexts that coincided with early-stage research into human genes throughout the past half century. We demonstrate that early-stage research into human genes differs in team size, citation impact, funding mechanisms, and publication outlet, but that generalized insights derived from studies of science and innovation only partially apply to early-stage research into human genes. Further, we demonstrate that, presently, genome biology accounts for most of the initial early-stage research, while subsequent early-stage research can engage other life sciences fields. We therefore anticipate that the specificity of our findings will enable scientists and policymakers to better promote early-stage research into human genes and increase overall innovation within the life sciences.

Aging is associated with a systemic length-associated transcriptome imbalance.

Aging is among the most important risk factors for morbidity and mortality. To contribute toward a molecular understanding of aging, we analyzed age-resolved transcriptomic data from multiple studies. Here, we show that transcript length alone explains most transcriptional changes observed with aging in mice and humans. We present three lines of evidence supporting the biological importance of the uncovered transcriptome imbalance. First, in vertebrates the length association primarily displays a lower relative abundance of long transcripts in aging. Second, eight antiaging interventions of the Interventions Testing Program of the National Institute on Aging can counter this length association. Third, we find that in humans and mice the genes with the longest transcripts enrich for genes reported to extend lifespan, whereas those with the shortest transcripts enrich for genes reported to shorten lifespan. Our study opens fundamental questions on aging and the organization of transcriptomes.

Spreader events and the limitations of projected networks for capturing dynamics on multipartite networks.

Many systems of scientific interest can be conceptualized as multipartite networks. Examples include the spread of sexually transmitted infections, scientific collaborations, human friendships, product recommendation systems, and metabolic networks. In practice, these systems are often studied after projection onto a single class of nodes, losing crucial information. Here, we address a significant knowledge gap by comparing transmission dynamics on temporal multipartite networks and on their time-aggregated unipartite projections to determine the impact of the lost information on our ability to predict the systems' dynamics. We show that the dynamics of transmission models can be dramatically dissimilar on multipartite networks and on their projections at three levels: final outcome, the magnitude of the variability from realization to realization, and overall shape of the temporal trajectory. We find that the ratio of the number of nodes to the number of active edges over the time-aggregation scale determines the ability of projected networks to capture the dynamics on the multipartite network. Finally, we explore which properties of a multipartite network are crucial in generating synthetic networks that better reproduce the dynamical behavior observed in real multipartite networks.

Conformational stability of the bacterial adhesin, FimH, with an inactivating mutation.

Allostery governing two conformational states is one of the proposed mechanisms for catch-bond behavior in adhesive proteins. In FimH, a catch-bond protein expressed by pathogenic bacteria, separation of two domains disrupts inhibition by the pilin domain. Thus, tensile force can induce a conformational change in the lectin domain, from an inactive state to an active state with high affinity. To better understand allosteric inhibition in two-domain FimH (H2 inactive), we use molecular dynamics simulations to study the lectin domain alone, which has high affinity (HL active), and also the lectin domain stabilized in the low-affinity conformation by an Arg-60-Pro mutation (HL mutant). Because ligand-binding induces an allostery-like conformational change in HL mutant, this more experimentally tractable version has been proposed as a "minimal model" for FimH. We find that HL mutant has larger backbone fluctuations than both H2 inactive and HL active, at the binding pocket and allosteric interdomain region. We use an internal coordinate system of dihedral angles to identify protein regions with differences in backbone and side chain dynamics beyond the putative allosteric pathway sites. By characterizing HL mutant dynamics for the first time, we provide additional insight into the transmission of allosteric information across the lectin domain and build upon structural and thermodynamic data in the literature to further support the use of HL mutant as a "minimal model." Understanding how to alter protein dynamics to prevent the allosteric conformational change may guide drug development to prevent infection by blocking FimH adhesion.

COVID-19 research risks ignoring important host genes due to pre-established research patterns.

It is known that research into human genes is heavily skewed towards genes that have been widely studied for decades, including many genes that were being studied before the productive phase of the Human Genome Project. This means that the genes most frequently investigated by the research community tend to be only marginally more important to human physiology and disease than a random selection of genes. Based on an analysis of 10,395 research publications about SARS-CoV-2 that mention at least one human gene, we report here that the COVID-19 literature up to mid-October 2020 follows a similar pattern. This means that a large number of host genes that have been implicated in SARS-CoV-2 infection by four genome-wide studies remain unstudied. While quantifying the consequences of this neglect is not possible, they could be significant.

Long-term patterns of gender imbalance in an industry without ability or level of interest differences.

Female representation has been slowly but steadily increasing in many sectors of society. One sector where one would expect to see gender parity is the movie industry, yet the representation of females in most functions within the U.S. movie industry remain surprisingly low. Here, we study the historical patterns of female representation among actors, directors, and producers in an attempt to gain insights into the possible causes of the lack of gender parity in the industry. Our analyses reveals a remarkable temporal coincidence between the collapse in female representation across all functions and the advent of the Studio System, a period when the major Hollywood studios controlled all aspects of the industry. Female representation among actors, directors, producers and writers dropped to extraordinarily low values during the emergence and consolidation of the Studio System that in some cases have not yet recovered to pre-Studio System levels. In order to explore some possible mechanisms behind these patterns, we investigate the association between the gender balance of actors, writers, directors, and producers and a number of economic indicators, movie industry indicators, and movie characteristics. We find robust, strong, and significant associations which are consistent with an important role for the gender of decision makers on the gender balance of other industry functions. While in no way demonstrating causality, our findings add new perspectives to the discussions of the reasons for female under-representation in fields such as computer science and medicine, that have also experienced dramatic changes in female representation.

Fly-QMA: Automated analysis of mosaic imaginal discs in Drosophila.

Mosaic analysis provides a means to probe developmental processes in situ by generating loss-of-function mutants within otherwise wildtype tissues. Combining these techniques with quantitative microscopy enables researchers to rigorously compare RNA or protein expression across the resultant clones. However, visual inspection of mosaic tissues remains common in the literature because quantification demands considerable labor and computational expertise. Practitioners must segment cell membranes or cell nuclei from a tissue and annotate the clones before their data are suitable for analysis. Here, we introduce Fly-QMA, a computational framework that automates each of these tasks for confocal microscopy images of Drosophila imaginal discs. The framework includes an unsupervised annotation algorithm that incorporates spatial context to inform the genetic identity of each cell. We use a combination of real and synthetic validation data to survey the performance of the annotation algorithm across a broad range of conditions. By contributing our framework to the open-source software ecosystem, we aim to contribute to the current move toward automated quantitative analysis among developmental biologists.

A quantitative approach for the analysis of clinician recognition of acute respiratory distress syndrome using electronic health record data.

IMPORTANCE: Despite its efficacy, low tidal volume ventilation (LTVV) remains severely underutilized for patients with acute respiratory distress syndrome (ARDS). Physician under-recognition of ARDS is a significant barrier to LTVV use. We propose a computational method that addresses some of the limitations of the current approaches to automated measurement of whether ARDS is recognized by physicians. OBJECTIVE: To quantify patient and physician factors affecting physicians' tidal volume selection and to build a computational model of physician recognition of ARDS that accounts for these factors. DESIGN, SETTING, AND PARTICIPANTS: In this cross-sectional study, electronic health record data were collected for 361 ARDS patients and 388 non-ARDS hypoxemic (control) patients in nine adult intensive care units at four hospitals between June 24 and December 31, 2013. METHODS: Standardized tidal volumes (mL/kg predicted body weight) were chosen as a proxy for physician decision-making behavior. Using data-science approaches, we quantified the effect of eight factors (six severity of illness, two physician behaviors) on selected standardized tidal volumes in ARDS and control patients. Significant factors were incorporated in computational behavioral models of physician recognition of ARDS. RESULTS: Hypoxemia severity and ARDS documentation in physicians' notes were associated with lower standardized tidal volumes in the ARDS cohort. Greater patient height was associated with lower standardized tidal volumes (which is already normalized for height) in both ARDS and control patients. The recognition model yielded a mean (99% confidence interval) physician recognition of ARDS of 22% (9%-42%) for mild, 34% (19%-49%) for moderate, and 67% (41%-100%) for severe ARDS. CONCLUSIONS AND RELEVANCE: In this study, patient characteristics and physician behaviors were demonstrated to be associated with differences in ventilator management in both ARDS and control patients. Our model of physician ARDS recognition measurement accounts for these clinical variables, providing an electronic approach that moves beyond relying on chart documentation or resource intensive approaches.