Comparison of Deep Learning Approaches for Conversion of International Classification of Diseases Codes to the Abbreviated Injury Scale.

The injury severity classifications generated from the Abbreviated Injury Scale (AIS) provide information that allows for standardized comparisons in the field of trauma injury research. However, the majority of injuries are coded in International Classification of Diseases (ICD) and lack this severity information. A system to predict injury severity classifications from ICD codes would be beneficial as manually coding in AIS can be time-intensive or even impossible for some retrospective cases. It has been previously shown that the encoder-decoder-based neural machine translation (NMT) model is more accurate than a one-to-one mapping of ICD codes to AIS. The objective of this study is to compare the accuracy of two architectures, feedforward neural networks (FFNN) and NMT, in predicting Injury Severity Score (ISS) and ISS ≥16 classification. Both architectures were tested in direct conversion from ICD codes to ISS score and indirect conversion through AIS for a total of four models. Trauma cases from the U.S. National Trauma Data Bank were used to develop and test the four models as the injuries were coded in both ICD and AIS. 2,031,793 trauma cases from 2017-2018 were used to train and validate the models while 1,091,792 cases from 2019 were used to test and compare them. The results showed that indirect conversion through AIS using an NMT was the most accurate in predicting the exact ISS score, followed by direct conversion with FFNN, direct conversion with NMT, and lastly indirect conversion with FFNN, with statistically significant differences in performance on all pairwise comparisons. The rankings were similar when comparing the accuracy of predicting ISS ≥16 classification, however the differences were smaller. The NMT architecture continues to demonstrate notable accuracy in predicting exact ISS scores, but a simpler FFNN approach may be preferred in specific situations, such as if only ISS ≥16 classification is needed or large-scale computational resources are unavailable.

Neighborhood Deprivation and Association With Neonatal Intensive Care Unit Mortality and Morbidity for Extremely Premature Infants.

Importance: Socioeconomic status affects pregnancy and neurodevelopment, but its association with hospital outcomes among premature infants is unknown. The Area Deprivation Index (ADI) is a validated measure of neighborhood disadvantage that uses US Census Bureau data on income, educational level, employment, and housing quality. Objective: To determine whether ADI is associated with neonatal intensive care unit (NICU) mortality and morbidity in extremely premature infants. Design, Setting, and Participants: This retrospective cohort study was performed at 4 level IV NICUs in the US Northeast, Mid-Atlantic, Midwest, and South regions. Non-Hispanic White and Black infants with gestational age of less than 29 weeks and born between January 1, 2012, and December 31, 2020, were included in the analysis. Addresses were converted to census blocks, identified by Federal Information Processing Series codes, to link residences to national ADI percentiles. Exposures: ADI, race, birth weight, sex, and outborn status. Main Outcomes and Measures: In the primary outcome, the association between ADI and NICU mortality was analyzed using bayesian logistic regression adjusted for race, birth weight, outborn status, and sex. Risk factors were considered significant if the 95% credible intervals excluded zero. In the secondary outcome, the association between ADI and NICU morbidities, including late-onset sepsis, necrotizing enterocolitis (NEC), and severe intraventricular hemorrhage (IVH), were also analyzed. Results: A total of 2765 infants with a mean (SD) gestational age of 25.6 (1.7) weeks and mean (SD) birth weight of 805 (241) g were included in the analysis. Of these, 1391 (50.3%) were boys, 1325 (47.9%) reported Black maternal race, 498 (18.0%) died before NICU discharge, 692 (25.0%) developed sepsis or NEC, and 353 (12.8%) had severe IVH. In univariate analysis, higher median ADI was found among Black compared with White infants (77 [IQR, 45-93] vs 57 [IQR, 32-77]; P < .001), those who died before NICU discharge vs survived (71 [IQR, 45-89] vs 64 [IQR, 36-86]), those with late-onset sepsis or NEC vs those without (68 [IQR, 41-88] vs 64 [IQR, 35-86]), and those with severe IVH vs those without (69 [IQR, 44-90] vs 64 [IQR, 36-86]). In a multivariable bayesian logistic regression model, lower birth weight, higher ADI, and male sex were risk factors for mortality (95% credible intervals excluded zero), while Black race and outborn status were not. The ADI was also identified as a risk factor for sepsis or NEC and severe IVH. Conclusions and Relevance: The findings of this cohort study of extremely preterm infants admitted to 4 NICUs in different US geographic regions suggest that ADI was a risk factor for mortality and morbidity after adjusting for multiple covariates.

Race, Class, and Place Modify Mortality Rates for the Leading Causes of Death in the United States, 1999-2021.

BACKGROUND: Race and ethnicity, socioeconomic class, and geographic location are well-known social determinants of health in the US. Studies of population mortality often consider two, but not all three of these risk factors. OBJECTIVES: To disarticulate the associations of race (whiteness), class (socioeconomic status), and place (county) with risk of cause-specific death in the US. DESIGN: We conducted a retrospective analysis of death certificate data. Bayesian regression models, adjusted for age and race/ethnicity from the American Community Survey and the county Area Deprivation Index, were used for inference. MAIN MEASURES: County-level mortality for 11 leading causes of death (1999-2019) and COVID-19 (2020-2021). KEY RESULTS: County "whiteness" and socioeconomic status modified death rates; geospatial effects differed by cause of death. Other factors equal, a 20% increase in county whiteness was associated with 5-8% increase in death from three causes and 4-15% reduction in death from others, including COVID-19. Other factors equal, advantaged counties had significantly lower death rates, even when juxtaposed with disadvantaged ones. Patterns of residual risk, measured by spatial county effects, varied by cause of death; for example: cancer and heart disease death rates were better explained by age, socioeconomic status, and county whiteness than were COVID-19 and suicide deaths. CONCLUSIONS: There are important independent contributions from race, class, and geography to risk of death in the US.

Polymer models reveal how chromatin modification can modulate force at the kinetochore.

A key feature of chromosome segregation is the ability to sense tension between sister kinetochores. DNA between sister kinetochores must be packaged in a way that sustains tension propagation from one kinetochore to its sister, approximately 1 micron away. A molecular bottlebrush consisting of a primary axis populated with a crowded array of side chains provides a means to build tension over length scales considerably larger than the stiffness of the individual elements, that is, DNA polymer. Evidence for the bottlebrush organization of chromatin between sister kinetochores comes from genetic, cell biological, and polymer modeling of the budding yeast centromere. In this study, we have used polymer dynamic simulations of the bottlebrush to recapitulate experimental observations of kinetochore structure. Several aspects of the spatial distribution of kinetochore proteins and their response to perturbation lack a mechanistic understanding. Changes in physical parameters of bottlebrush, DNA stiffness, and DNA loops directly impact the architecture of the inner kinetochore. This study reveals that the bottlebrush is an active participant in building tension between sister kinetochores and proposes a mechanism for chromatin feedback to the kinetochore.

AI-Assisted Forward Modeling of Biological Structures.

The rise of machine learning and deep learning technologies have allowed researchers to automate image classification. We describe a method that incorporates automated image classification and principal component analysis to evaluate computational models of biological structures. We use a computational model of the kinetochore to demonstrate our artificial-intelligence (AI)-assisted modeling method. The kinetochore is a large protein complex that connects chromosomes to the mitotic spindle to facilitate proper cell division. The kinetochore can be divided into two regions: the inner kinetochore, including proteins that interact with DNA; and the outer kinetochore, comprised of microtubule-binding proteins. These two kinetochore regions have been shown to have different distributions during metaphase in live budding yeast and therefore act as a test case for our forward modeling technique. We find that a simple convolutional neural net (CNN) can correctly classify fluorescent images of inner and outer kinetochore proteins and show a CNN trained on simulated, fluorescent images can detect difference in experimental images. A polymer model of the ribosomal DNA locus serves as a second test for the method. The nucleolus surrounds the ribosomal DNA locus and appears amorphous in live-cell, fluorescent microscopy experiments in budding yeast, making detection of morphological changes challenging. We show a simple CNN can detect subtle differences in simulated images of the ribosomal DNA locus, demonstrating our CNN-based classification technique can be used on a variety of biological structures.

Geometric partitioning of cohesin and condensin is a consequence of chromatin loops.

SMC (structural maintenance of chromosomes) complexes condensin and cohesin are crucial for proper chromosome organization. Condensin has been reported to be a mechanochemical motor capable of forming chromatin loops, while cohesin passively diffuses along chromatin to tether sister chromatids. In budding yeast, the pericentric region is enriched in both condensin and cohesin. As in higher-eukaryotic chromosomes, condensin is localized to the axial chromatin of the pericentric region, while cohesin is enriched in the radial chromatin. Thus, the pericentric region serves as an ideal model for deducing the role of SMC complexes in chromosome organization. We find condensin-mediated chromatin loops establish a robust chromatin organization, while cohesin limits the area that chromatin loops can explore. Upon biorientation, extensional force from the mitotic spindle aggregates condensin-bound chromatin from its equilibrium position to the axial core of pericentric chromatin, resulting in amplified axial tension. The axial localization of condensin depends on condensin's ability to bind to chromatin to form loops, while the radial localization of cohesin depends on cohesin's ability to diffuse along chromatin. The different chromatin-tethering modalities of condensin and cohesin result in their geometric partitioning in the presence of an extensional force on chromatin.

RotoStep: A Chromosome Dynamics Simulator Reveals Mechanisms of Loop Extrusion.

ChromoShake is a three-dimensional simulator designed to explore the range of configurational states a chromosome can adopt based on thermodynamic fluctuations of the polymer chain. Here, we refine ChromoShake to generate dynamic simulations of a DNA-based motor protein such as condensin walking along the chromatin substrate. We model walking as a rotation of DNA-binding heat-repeat proteins around one another. The simulation is applied to several configurations of DNA to reveal the consequences of mechanical stepping on taut chromatin under tension versus loop extrusion on single-tethered, floppy chromatin substrates. These simulations provide testable hypotheses for condensin and other DNA-based motors functioning along interphase chromosomes. Our model reveals a novel mechanism for condensin enrichment in the pericentromeric region of mitotic chromosomes. Increased condensin dwell time at centromeres results in a high density of pericentric loops that in turn provide substrate for additional condensin.