Search for: All records

This paper studies the performance of large language models (LLMs), particularly regarding demographic fairness, in solving real-world healthcare tasks. We evaluate state-of-the-art LLMs with three prevalent learning frameworks across six diverse healthcare tasks and find significant challenges in applying LLMs to real-world healthcare tasks and persistent fairness issues across demographic groups. We also find that explicitly providing demographic information yields mixed results, while LLM`s ability to infer such details raises concerns about biased health predictions. Utilizing LLMs as autonomous agents with access to up-to-date guidelines does not guarantee performance improvement. We believe these findings reveal the critical limitations of LLMs in healthcare fairness and the urgent need for specialized research in this area. 

This paper studies the performance of large language models (LLMs), particularly regarding demographic fairness, in solving real-world healthcare tasks. We evaluate state-of-the-art LLMs with three prevalent learning frameworks across six diverse healthcare tasks and find significant challenges in applying LLMs to real-world healthcare tasks and persistent fairness issues across demographic groups. We also find that explicitly providing demographic information yields mixed results, while LLM’s ability to infer such details raises concerns about biased health predictions. Utilizing LLMs as autonomous agents with access to up-to-date guidelines does not guarantee performance improvement. We believe these findings reveal the critical limitations of LLMs in healthcare fairness and the urgent need for specialized research in this area. 

Health coaching helps patients achieve personalized and lifestyle-related goals, effectively managing chronic conditions and alleviating mental health issues. It is particularly beneficial, however cost-prohibitive, for low-socioeconomic status populations due to its highly personalized and labor-intensive nature. In this paper, we propose a neuro-symbolic goal summarizer to support health coaches in keeping track of the goals and a text-units-text dialogue generation model that converses with patients and helps them create and accomplish specific goals for physical activities. Our models outperform previous state-of-the-art while eliminating the need for predefined schema and corresponding annotation. We also propose a new health coaching dataset extending previous work and a metric to measure the unconventionality of the patient’s response based on data difficulty, facilitating potential coach alerts during deployment. 

Abstract Cadmium sulfide (CdS) pigments have degraded in several well-known artworks, but the influence of pigment properties and environmental conditions on the degradation process have yet to be fully understood. Traditional non-destructive analysis techniques primarily focus on macroscopic degradation, whereas microscopic information is typically obtained with invasive techniques that require sample removal. Here, we demonstrate the use of pump-probe microscopy to nondestructively visualize the three-dimensional structure and degradation progress of CdS pigments in oil paints. CdS pigments, reproduced following historical synthesis methods, were reproduced as oil paints and artificially aged by exposure to high relative humidity and light. The degradation of CdS to CdSO₄·xH₂O was confirmed by both FTIR (Fourier-transform infrared) and XPS (x-ray photoelectron spectroscopy) experiments. During the degradation process, optical pump-probe microscopy was applied to track the degradation progress in single grains, and volumetric imaging revealed early CdS degradation of small particles and on the surface of large particles. This indicates that the particle dimension influences the extent and evolution of degradation of historical CdS. In addition, the pump-probe signal decrease in degraded CdS is observable before visible changes to the eye, demonstrating that pump-probe microscopy is a promising tool to detect early-stage degradation in artworks. 

ABSTRACT Systemic infections byCandidaspp. are associated with high mortality rates, partly due to limitations in current antifungals, highlighting the need for novel drugs and drug targets. The fungal phosphatidylserine synthase, Cho1, fromCandida albicansis a logical antifungal drug target due to its importance in virulence, absence in the host, and conservation among fungal pathogens. Inhibitors of Cho1 could serve as lead compounds for drug development, so we developed a target-based screen for inhibitors of purified Cho1. This enzyme condenses serine and cytidyldiphosphate-diacylglycerol (CDP-DAG) into phosphatidylserine (PS) and releases cytidylmonophosphate (CMP). Accordingly, we developed anin vitronucleotidase-coupled malachite-green-based high throughput assay for purifiedC. albicansCho1 that monitors CMP production as a proxy for PS synthesis. Over 7,300 molecules curated from repurposing chemical libraries were interrogated in primary and dose-responsivity assays using this platform. The screen had a promising averageZ’ score of ~0.8, and seven compounds were identified that inhibit Cho1. Three of these, ebselen, LOC14, and CBR-5884, exhibited antifungal effects againstC. albicanscells, with fungicidal inhibition by ebselen and fungistatic inhibition by LOC14 and CBR-5884. Only CBR-5884 showed evidence of disruptingin vivoCho1 function by inducing phenotypes consistent with thecho1∆∆mutant, including a reduction of cellular PS levels. Kinetics curves and computational docking indicate that CBR-5884 competes with serine for binding to Cho1 with aK_iof 1,550 ± 245.6 nM. Thus, this compound has the potential for development into an antifungal compound. IMPORTANCEFungal phosphatidylserine synthase (Cho1) is a logical antifungal target due to its crucial role in the virulence and viability of various fungal pathogens, and since it is absent in humans, drugs targeted at Cho1 are less likely to cause toxicity in patients. Using fungal Cho1 as a model, there have been two unsuccessful attempts to discover inhibitors for Cho1 homologs in whole-cell screens prior to this study. The compounds identified in these attempts do not act directly on the protein, resulting in the absence of known Cho1 inhibitors. The significance of our research is that we developed a high-throughput target-based assay and identified the first Cho1 inhibitor, CBR-5884, which acts both on the purified protein and its function in the cell. This molecule acts as a competitive inhibitor with aK_ivalue of 1,550 ± 245.6 nM and, thus, has the potential for development into a new class of antifungals targeting PS synthase. 

The shared real-time information about natural disasters on social media platforms like Twitter and Facebook plays a critical role in informing volunteers, emergency managers, and response organizations. However, supervised learning models for monitoring disaster events require large amounts of annotated data, making them unrealistic for real-time use in disaster events. To address this challenge, we present a fine-grained disaster tweet classification model under the semi-supervised, few-shot learning setting where only a small number of annotated data is required. Our model, CrisisMatch, effectively classifies tweets into fine-grained classes of interest using few labeled data and large amounts of unlabeled data, mimicking the early stage of a disaster. Through integrating effective semi-supervised learning ideas and incorporating TextMixUp, CrisisMatch achieves performance improvement on two disaster datasets of 11.2% on average. Further analyses are also provided for the influence of the number of labeled data and out-of-domain results. 

Abstract Constructing an artificial solid electrolyte interphase (SEI) on lithium metal electrodes is a promising approach to address the rampant growth of dangerous lithium morphologies (dendritic and dead Li⁰) and low Coulombic efficiency that plague development of lithium metal batteries, but how Li⁺transport behavior in the SEI is coupled with mechanical properties remains unknown. We demonstrate here a facile and scalable solution-processed approach to form a Li₃N-rich SEI with a phase-pure crystalline structure that minimizes the diffusion energy barrier of Li⁺across the SEI. Compared with a polycrystalline Li₃N SEI obtained from conventional practice, the phase-pure/single crystalline Li₃N-rich SEI constitutes an interphase of high mechanical strength and low Li⁺diffusion barrier. We elucidate the correlation among Li⁺transference number, diffusion behavior, concentration gradient, and the stability of the lithium metal electrode by integrating phase field simulations with experiments. We demonstrate improved reversibility and charge/discharge cycling behaviors for both symmetric cells and full lithium-metal batteries constructed with this Li₃N-rich SEI. These studies may cast new insight into the design and engineering of an ideal artificial SEI for stable and high-performance lithium metal batteries.