Mechanistically Guided Materials Chemistry: Synthesis of Ternary Nitrides, CaZrN2 and CaHfN2.

Recent computational studies have predicted many new ternary nitrides, revealing synthetic opportunities in this underexplored phase space. However, synthesizing new ternary nitrides is difficult, in part because intermediate and product phases often have high cohesive energies that inhibit diffusion. Here, we report the synthesis of two new phases, calcium zirconium nitride (CaZrN2) and calcium hafnium nitride (CaHfN2), by solid state metathesis reactions between Ca3N2 and MCl4 (M = Zr, Hf). Although the reaction nominally proceeds to the target phases in a 1:1 ratio of the precursors via Ca3N2 + MCl4 → CaMN2 + 2 CaCl2, reactions prepared this way result in Ca-poor materials (CaxM2-xN2, x < 1). A small excess of Ca3N2 (ca. 20 mol %) is needed to yield stoichiometric CaMN2, as confirmed by high-resolution synchrotron powder X-ray diffraction. In situ synchrotron X-ray diffraction studies reveal that nominally stoichiometric reactions produce Zr3+ intermediates early in the reaction pathway, and the excess Ca3N2 is needed to reoxidize Zr3+ intermediates back to the Zr4+ oxidation state of CaZrN2. Analysis of computationally derived chemical potential diagrams rationalizes this synthetic approach and its contrast from the synthesis of MgZrN2. These findings additionally highlight the utility of in situ diffraction studies and computational thermochemistry to provide mechanistic guidance for synthesis.

Using machine learning to develop smart reflex testing protocols.

OBJECTIVE: Reflex testing protocols allow clinical laboratories to perform second line diagnostic tests on existing specimens based on the results of initially ordered tests. Reflex testing can support optimal clinical laboratory test ordering and diagnosis. In current clinical practice, reflex testing typically relies on simple "if-then" rules; however, this limits the opportunities for reflex testing since most test ordering decisions involve more complexity than traditional rule-based approaches would allow. Here, using the analyte ferritin as an example, we propose an alternative machine learning-based approach to "smart" reflex testing. METHODS: Using deidentified patient data, we developed a machine learning model to predict whether a patient getting CBC testing will also have ferritin testing ordered. We evaluate applications of this model to reflex testing by assessing its performance in comparison to possible rule-based approaches. RESULTS: Our underlying machine learning models performed moderately well in predicting ferritin test ordering (AUC=0.731 in reference to actual ordering) and demonstrated promising potential to underlie key clinical applications. In contrast, none of the many traditionally framed, rule-based, hypothetical reflex protocols we evaluated offered sufficient agreement with actual ordering to be clinically feasible. Using chart review, we further demonstrated that the strategic deployment of our model could avoid important ferritin test ordering errors. CONCLUSIONS: Machine learning may provide a foundation for new types of reflex testing with enhanced benefits for clinical diagnosis.

Selective Synthesis of Defect-Rich LaMnO3 by Low-Temperature Anion Cometathesis.

The synthesis of complex oxides at low temperatures brings forward aspects of chemistry not typically considered. This study focuses on perovskite LaMnO3, which is of interest for its correlated electronic behavior tied to the oxidation state and thus the spin configuration of manganese. Traditional equilibrium synthesis of these materials typically requires synthesis reaction temperatures in excess of 1000 °C, followed by subsequent annealing steps at lower temperatures and different p(O2) conditions to manipulate the oxygen content postsynthesis (e.g., LaMnO3+x). Double-ion exchange (metathesis) reactions have recently been shown to react at much lower temperatures (500-800 °C), highlighting a fundamental knowledge gap for how solids react at lower temperatures. Here, we revisit the metathesis reaction, LiMnO2 + LaOX, where X is a halide or mixture of halides, using in situ synchrotron X-ray diffraction. These experiments reveal low reaction onset temperatures (ca. 450-480 °C). The lowest reaction temperatures are achieved by a mixture of lanthanum oxyhalide precursors: 2 LiMnO2 + LaOCl + LaOBr. In all cases, the resulting products are the expected alkali halide salt and defective La1-ϵMn1-ϵO3, where ϵ = x/(3 + x). We observe a systematic variation in defect concentration, consistent with a rapid stoichiometric local equilibration of the precursors and the subsequent global thermodynamic equilibration with O2 (g), as revealed by computational thermodynamics. Together, these results reveal how the inclusion of additional elements (e.g., Li and a halide) leads to the local equilibrium, particularly at low reaction temperatures for solid-state chemistry.

An autonomous laboratory for the accelerated synthesis of novel materials.

To close the gap between the rates of computational screening and experimental realization of novel materials1,2, we introduce the A-Lab, an autonomous laboratory for the solid-state synthesis of inorganic powders. This platform uses computations, historical data from the literature, machine learning (ML) and active learning to plan and interpret the outcomes of experiments performed using robotics. Over 17 days of continuous operation, the A-Lab realized 41 novel compounds from a set of 58 targets including a variety of oxides and phosphates that were identified using large-scale ab initio phase-stability data from the Materials Project and Google DeepMind. Synthesis recipes were proposed by natural-language models trained on the literature and optimized using an active-learning approach grounded in thermodynamics. Analysis of the failed syntheses provides direct and actionable suggestions to improve current techniques for materials screening and synthesis design. The high success rate demonstrates the effectiveness of artificial-intelligence-driven platforms for autonomous materials discovery and motivates further integration of computations, historical knowledge and robotics.

Assessing Thermodynamic Selectivity of Solid-State Reactions for the Predictive Synthesis of Inorganic Materials.

Synthesis is a major challenge in the discovery of new inorganic materials. Currently, there is limited theoretical guidance for identifying optimal solid-state synthesis procedures. We introduce two selectivity metrics, primary and secondary competition, to assess the favorability of target/impurity phase formation in solid-state reactions. We used these metrics to analyze 3520 solid-state reactions in the literature, ranking existing approaches to popular target materials. Additionally, we implemented these metrics in a data-driven synthesis planning workflow and demonstrated its application in the synthesis of barium titanate (BaTiO3). Using an 18-element chemical reaction network with first-principles thermodynamic data from the Materials Project, we identified 82985 possible BaTiO3 synthesis reactions and selected 9 for experimental testing. Characterization of reaction pathways via synchrotron powder X-ray diffraction reveals that our selectivity metrics correlate with observed target/impurity formation. We discovered two efficient reactions using unconventional precursors (BaS/BaCl2 and Na2TiO3) that produce BaTiO3 faster and with fewer impurities than conventional methods, highlighting the importance of considering complex chemistries with additional elements during precursor selection. Our framework provides a foundation for predictive inorganic synthesis, facilitating the optimization of existing recipes and the discovery of new materials, including those not easily attainable with conventional precursors.

Single-cell mapping identifies MSI+ cells as a common origin for diverse subtypes of pancreatic cancer.

Identifying the cells from which cancers arise is critical for understanding the molecular underpinnings of tumor evolution. To determine whether stem/progenitor cells can serve as cells of origin, we created a Msi2-CreERT2 knock-in mouse. When crossed to CAG-LSL-MycT58A mice, Msi2-CreERT2 mice developed multiple pancreatic cancer subtypes: ductal, acinar, adenosquamous, and rare anaplastic tumors. Combining single-cell genomics with computational analysis of developmental states and lineage trajectories, we demonstrate that MYC preferentially triggers transformation of the most immature MSI2+ pancreas cells into multi-lineage pre-cancer cells. These pre-cancer cells subsequently diverge to establish pancreatic cancer subtypes by activating distinct transcriptional programs and large-scale genomic changes, and enforced expression of specific signals like Ras can redirect subtype specification. This study shows that multiple pancreatic cancer subtypes can arise from a common pool of MSI2+ cells and provides a powerful model to understand and control the programs that shape divergent fates in pancreatic cancer.

APPRAISE-AI Tool for Quantitative Evaluation of AI Studies for Clinical Decision Support.

Importance: Artificial intelligence (AI) has gained considerable attention in health care, yet concerns have been raised around appropriate methods and fairness. Current AI reporting guidelines do not provide a means of quantifying overall quality of AI research, limiting their ability to compare models addressing the same clinical question. Objective: To develop a tool (APPRAISE-AI) to evaluate the methodological and reporting quality of AI prediction models for clinical decision support. Design, Setting, and Participants: This quality improvement study evaluated AI studies in the model development, silent, and clinical trial phases using the APPRAISE-AI tool, a quantitative method for evaluating quality of AI studies across 6 domains: clinical relevance, data quality, methodological conduct, robustness of results, reporting quality, and reproducibility. These domains included 24 items with a maximum overall score of 100 points. Points were assigned to each item, with higher points indicating stronger methodological or reporting quality. The tool was applied to a systematic review on machine learning to estimate sepsis that included articles published until September 13, 2019. Data analysis was performed from September to December 2022. Main Outcomes and Measures: The primary outcomes were interrater and intrarater reliability and the correlation between APPRAISE-AI scores and expert scores, 3-year citation rate, number of Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) low risk-of-bias domains, and overall adherence to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement. Results: A total of 28 studies were included. Overall APPRAISE-AI scores ranged from 33 (low quality) to 67 (high quality). Most studies were moderate quality. The 5 lowest scoring items included source of data, sample size calculation, bias assessment, error analysis, and transparency. Overall APPRAISE-AI scores were associated with expert scores (Spearman ρ, 0.82; 95% CI, 0.64-0.91; P < .001), 3-year citation rate (Spearman ρ, 0.69; 95% CI, 0.43-0.85; P < .001), number of QUADAS-2 low risk-of-bias domains (Spearman ρ, 0.56; 95% CI, 0.24-0.77; P = .002), and adherence to the TRIPOD statement (Spearman ρ, 0.87; 95% CI, 0.73-0.94; P < .001). Intraclass correlation coefficient ranges for interrater and intrarater reliability were 0.74 to 1.00 for individual items, 0.81 to 0.99 for individual domains, and 0.91 to 0.98 for overall scores. Conclusions and Relevance: In this quality improvement study, APPRAISE-AI demonstrated strong interrater and intrarater reliability and correlated well with several study quality measures. This tool may provide a quantitative approach for investigators, reviewers, editors, and funding organizations to compare the research quality across AI studies for clinical decision support.

Direct Mechanochemical Synthesis, Phase Stability, and Electrochemical Performance of α-NaFeO2.

To better understand polymorph control in transition metal oxides, the mechanochemical synthesis of NaFeO2 was explored. Herein, we report the direct synthesis of α-NaFeO2 through a mechanochemical process. By milling Na2O2 and γ-Fe2O3 for 5 h, α-NaFeO2 was prepared without high-temperature annealing needed in other synthesis methods. While investigating the mechanochemical synthesis, it was observed that changing the starting precursors and mass of precursors affects the resulting NaFeO2 structure. Density functional theory calculations on the phase stability of NaFeO2 phases show that the α phase is stabilized over the ß phase in oxidizing environments, which is provided by the oxygen-rich reaction between Na2O2 and Fe2O3. This provides a possible route to understanding polymorph control in NaFeO2. Annealing the as-milled α-NaFeO2 at 700 °C has resulted in increased crystallinity and structural changes that improved electrochemical performance in terms of capacity over the as-milled sample.

Using Machine Learning to Develop Smart Reflex Testing Protocols.

Objective: Reflex testing protocols allow clinical laboratories to perform second line diagnostic tests on existing specimens based on the results of initially ordered tests. Reflex testing can support optimal clinical laboratory test ordering and diagnosis. In current clinical practice, reflex testing typically relies on simple "if-then" rules; however, this limits their scope since most test ordering decisions involve more complexity than a simple rule will allow. Here, using the analyte ferritin as an example, we propose an alternative machine learning-based approach to "smart" reflex testing with a wider scope and greater impact than traditional rule-based approaches. Methods: Using patient data, we developed a machine learning model to predict whether a patient getting CBC testing will also have ferritin testing ordered, consider applications of this model to "smart" reflex testing, and evaluate the model by comparing its performance to possible rule-based approaches. Results: Our underlying machine learning models performed moderately well in predicting ferritin test ordering and demonstrated greater suitability to reflex testing than rule-based approaches. Using chart review, we demonstrate that our model may improve ferritin test ordering. Finally, as a secondary goal, we demonstrate that ferritin test results are missing not at random (MNAR), a finding with implications for unbiased imputation of missing test results. Conclusions: Machine learning may provide a foundation for new types of reflex testing with enhanced benefits for clinical diagnosis and laboratory utilization management.

Clinical Artificial Intelligence: Design Principles and Fallacies.

Clinical artificial intelligence (AI)/machine learning (ML) is anticipated to offer new abilities in clinical decision support, diagnostic reasoning, precision medicine, clinical operational support, and clinical research, but careful concern is needed to ensure these technologies work effectively in the clinic. Here, we detail the clinical ML/AI design process, identifying several key questions and detailing several common forms of issues that arise with ML tools, as motivated by real-world examples, such that clinicians and researchers can better anticipate and correct for such issues in their own use of ML/AI techniques.

Osteoporosis-related characteristics in care home residents in England: a retrospective cohort study.

BACKGROUND: The characteristics of care home populations, with respect to fracture risk factors, have not been well-defined. AIM: To describe osteoporosis-related characteristics among care home residents, including fracture risk factors, fracture rates, post-fracture outcomes, and osteoporosis treatment duration. DESIGN & SETTING: A descriptive cohort study of care home residents aged ≥60 years (n = 8366) and a matched cohort of non-care home residents (n = 16 143) in England from 2012 to 2019. Clinical Practice Research Datalink (CPRD) linked to Hospital Episode Statistics (HES) and Office for National Statistics (ONS) death data were used. METHOD: The characteristics were assessed using descriptive statistics. Fracture risk factors and fracture rates were described in both the care home and matched population. In the care home population, Kaplan-Meier curves were plotted to assess osteoporosis treatment duration. RESULTS: At index, fracture risk factors were more common in care home residents versus the matched cohort, including body mass index (BMI) <18.5 (12.2% versus 5.1%), history of falls (48.9% versus 30.7%), prior fracture (26.5% versus 10.8%), and prior hip fracture (17.1% versus 5.8%). Fracture rate was 43.5 (95% confidence interval [CI] = 39.7 to 47.5) in care home residents and 28.0 (95% CI = 26.3 to 29.9) per 1000 person-years in the matched cohort. Overall, osteoporosis treatment was initiated in 3.6% (n = 225/6265) of care home residents and 45.9% remained on treatment at 12 months. Among care home residents who experienced fracture, 21.9% (n = 72/329) received an osteoporosis diagnosis; 21.2% (n = 63/297) initiated osteoporosis treatment post-hip fracture. CONCLUSION: Care home residents had more fracture risk factors and higher fracture rates than the matched cohort; however, osteoporosis diagnosis, treatment rates, and treatment duration were low. There is an opportunity to improve osteoporosis management in this vulnerable population.

Smarcd3 is an epigenetic modulator of the metabolic landscape in pancreatic ductal adenocarcinoma.

Pancreatic cancer is characterized by extensive resistance to conventional therapies, making clinical management a challenge. Here we map the epigenetic dependencies of cancer stem cells, cells that preferentially evade therapy and drive progression, and identify SWI/SNF complex member SMARCD3 as a regulator of pancreatic cancer cells. Although SWI/SNF subunits often act as tumor suppressors, we show that SMARCD3 is amplified in cancer, enriched in pancreatic cancer stem cells and upregulated in the human disease. Diverse genetic mouse models of pancreatic cancer and stage-specific Smarcd3 deletion reveal that Smarcd3 loss preferentially impacts established tumors, improving survival especially in context of chemotherapy. Mechanistically, SMARCD3 acts with FOXA1 to control lipid and fatty acid metabolism, programs associated with therapy resistance and poor prognosis in cancer. These data identify SMARCD3 as an epigenetic modulator responsible for establishing the metabolic landscape in aggressive pancreatic cancer cells and a potential target for new therapies.

Chemical reaction networks and opportunities for machine learning.

Chemical reaction networks (CRNs), defined by sets of species and possible reactions between them, are widely used to interrogate chemical systems. To capture increasingly complex phenomena, CRNs can be leveraged alongside data-driven methods and machine learning (ML). In this Perspective, we assess the diverse strategies available for CRN construction and analysis in pursuit of a wide range of scientific goals, discuss ML techniques currently being applied to CRNs and outline future CRN-ML approaches, presenting scientific and technical challenges to overcome.

Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations.

Artificial intelligence (AI) systems have increasingly achieved expert-level performance in medical imaging applications. However, there is growing concern that such AI systems may reflect and amplify human bias, and reduce the quality of their performance in historically under-served populations such as female patients, Black patients, or patients of low socioeconomic status. Such biases are especially troubling in the context of underdiagnosis, whereby the AI algorithm would inaccurately label an individual with a disease as healthy, potentially delaying access to care. Here, we examine algorithmic underdiagnosis in chest X-ray pathology classification across three large chest X-ray datasets, as well as one multi-source dataset. We find that classifiers produced using state-of-the-art computer vision techniques consistently and selectively underdiagnosed under-served patient populations and that the underdiagnosis rate was higher for intersectional under-served subpopulations, for example, Hispanic female patients. Deployment of AI systems using medical imaging for disease diagnosis with such biases risks exacerbation of existing care biases and can potentially lead to unequal access to medical treatment, thereby raising ethical concerns for the use of these models in the clinic.

Selectivity in Yttrium Manganese Oxide Synthesis via Local Chemical Potentials in Hyperdimensional Phase Space.

In sharp contrast to molecular synthesis, materials synthesis is generally presumed to lack selectivity. The few known methods of designing selectivity in solid-state reactions have limited scope, such as topotactic reactions or strain stabilization. This contribution describes a general approach for searching large chemical spaces to identify selective reactions. This novel approach explains the ability of a nominally "innocent" Na2CO3 precursor to enable the metathesis synthesis of single-phase Y2Mn2O7: an outcome that was previously only accomplished at extreme pressures and which cannot be achieved with closely related precursors of Li2CO3 and K2CO3 under identical conditions. By calculating the required change in chemical potential across all possible reactant-product interfaces in an expanded chemical space including Y, Mn, O, alkali metals, and halogens, using thermodynamic parameters obtained from density functional theory calculations, we identify reactions that minimize the thermodynamic competition from intermediates. In this manner, only the Na-based intermediates minimize the distance in the hyperdimensional chemical potential space to Y2Mn2O7, thus providing selective access to a phase which was previously thought to be metastable. Experimental evidence validating this mechanism for pathway-dependent selectivity is provided by intermediates identified from in situ synchrotron-based crystallographic analysis. This approach of calculating chemical potential distances in hyperdimensional compositional spaces provides a general method for designing selective solid-state syntheses that will be useful for gaining access to metastable phases and for identifying reaction pathways that can reduce the synthesis temperature, and cost, of technological materials.

A framework for quantifying uncertainty in DFT energy corrections.

In this work, we demonstrate a method to quantify uncertainty in corrections to density functional theory (DFT) energies based on empirical results. Such corrections are commonly used to improve the accuracy of computational enthalpies of formation, phase stability predictions, and other energy-derived properties, for example. We incorporate this method into a new DFT energy correction scheme comprising a mixture of oxidation-state and composition-dependent corrections and show that many chemical systems contain unstable polymorphs that may actually be predicted stable when uncertainty is taken into account. We then illustrate how these uncertainties can be used to estimate the probability that a compound is stable on a compositional phase diagram, thus enabling better-informed assessments of compound stability.

A graph-based network for predicting chemical reaction pathways in solid-state materials synthesis.

Accelerated inorganic synthesis remains a significant challenge in the search for novel, functional materials. Many of the principles which enable "synthesis by design" in synthetic organic chemistry do not exist in solid-state chemistry, despite the availability of extensive computed/experimental thermochemistry data. In this work, we present a chemical reaction network model for solid-state synthesis constructed from available thermochemistry data and devise a computationally tractable approach for suggesting likely reaction pathways via the application of pathfinding algorithms and linear combination of lowest-cost paths in the network. We demonstrate initial success of the network in predicting complex reaction pathways comparable to those reported in the literature for YMnO3, Y2Mn2O7, Fe2SiS4, and YBa2Cu3O6.5. The reaction network presents opportunities for enabling reaction pathway prediction, rapid iteration between experimental/theoretical results, and ultimately, control of the synthesis of solid-state materials.

Reproducibility in machine learning for health research: Still a ways to go.

Machine learning for health must be reproducible to ensure reliable clinical use. We evaluated 511 scientific papers across several machine learning subfields and found that machine learning for health compared poorly to other areas regarding reproducibility metrics, such as dataset and code accessibility. We propose recommendations to address this problem.

CheXclusion: Fairness gaps in deep chest X-ray classifiers.

Machine learning systems have received much attention recently for their ability to achieve expert-level performance on clinical tasks, particularly in medical imaging. Here, we examine the extent to which state-of-the-art deep learning classifiers trained to yield diagnostic labels from X-ray images are biased with respect to protected attributes. We train convolution neural networks to predict 14 diagnostic labels in 3 prominent public chest X-ray datasets: MIMIC-CXR, Chest-Xray8, CheXpert, as well as a multi-site aggregation of all those datasets. We evaluate the TPR disparity - the difference in true positive rates (TPR) - among different protected attributes such as patient sex, age, race, and insurance type as a proxy for socioeconomic status. We demonstrate that TPR disparities exist in the state-of-the-art classifiers in all datasets, for all clinical tasks, and all subgroups. A multi-source dataset corresponds to the smallest disparities, suggesting one way to reduce bias. We find that TPR disparities are not significantly correlated with a subgroup's proportional disease burden. As clinical models move from papers to products, we encourage clinical decision makers to carefully audit for algorithmic disparities prior to deployment. Our supplementary materials can be found at, http://www.marzyehghassemi.com/chexclusion-supp-3/.