Citations

Abstract Objective The fast accumulation of vast pharmacogenomics data of cancer cell lines provide unprecedented opportunities for drug sensitivity prediction (DSP), a crucial prerequisite for the advancement of precision oncology. Recently, Generative Large Language Models (LLM) have demonstrated performance and generalization prowess across diverse tasks in the field of natural language processing (NLP). However, the structured format of the pharmacogenomics data poses challenge for the utility of LLM in DSP. Therefore, the objective of this study is multi-fold: to adapt prompt engineering for structured pharmacogenomics data toward optimizing LLM’s DSP performance, to evaluate LLM’s generalization in real-world DSP scenarios, and to compare LLM’s DSP performance against that of state-of-the-science baselines. Methods We systematically investigated the capability of the Generative Pre-trained Transformer (GPT) as a DSP model on four publicly available benchmark pharmacogenomics datasets, which are stratified by five cancer tissue types of cell lines and encompass both oncology and non-oncology drugs. Essentially, the predictive landscape of GPT is assessed for effectiveness on the DSP task via four learning paradigms: zero-shot learning, few-shot learning, fine-tuning and clustering pretrained embeddings. To facilitate GPT in seamlessly processing the structured pharmacogenomics data, domain-specific novel prompt engineering is employed by implementing three prompt templates (i.e., Instruction, Instruction-Prefix, Cloze) and integrating pharmacogenomics-related features into the prompt. We validated GPT’s performance in diverse real-world DSP scenarios: cross-tissue generalization, blind tests, and analyses of drug-pathway associations and top sensitive/resistant cell lines. Furthermore, we conducted a comparative evaluation of GPT against multiple Transformer-based pretrained models and existing DSP baselines. Results Extensive experiments on the pharmacogenomics datasets across the five tissue cohorts demonstrate that fine-tuning GPT yields the best DSP performance (28% F1 increase, p-value= 0.0003) followed by clustering pretrained GPT embeddings (26% F1 increase, p-value= 0.0005), outperforming GPT in-context learning (i.e., few-shot). However, GPT in the zero-shot setting had a big F1 gap, resulting in the worst performance. Within the scope of prompt engineering, performance enhancement was achieved by directly instructing GPT about the DSP task and resorting to a concise context format (i.e., instruction-prefix), leading to F1 performance gain of 22% (p-value=0.02); while incorporation of drug-cell line prompt context derived from genomics and/or molecular features further boosted F1 score by 2%. Compared to state-of-the-science DSP baselines, GPT significantly asserted superior mean F1 performance (16% gain, p-value<0.05) on the GDSC dataset. In the crosstissue analysis, GPT showcased comparable generalizability to the within-tissue performances on the GDSC and PRISM datasets, while statistically significant F1 performance improvements on the CCLE (8%, p-value=0.001) and DrugComb (19%, p-value=0.009) datasets. Evaluation on the challenging blind tests suggests GPT’s competitiveness on the CCLE and DrugComb datasets compared to random splitting. Furthermore, analyses of the drug-pathway associations and log probabilities provided valuable insights that align with previous DSP findings. Conclusion The diverse experiment setups and in-depth analysis underscore the importance of generative LLM, such as GPT, as a viable in silico approach to guide precision oncology. Availability https://github.com/bioIKEA/SensitiveCancerGPT

BACKGROUND: Catheter ablation of ventricular arrhythmias, one of the most rapidly growing procedures in cardiac electrophysiology, is associated with magnetic resonance imaging–detected brain lesions in more than half of cases. Although a retrograde aortic approach is conventional, modern tools enable entry through a transseptal approach that may avoid embolization of debris from the arterial system. We sought to test the hypothesis that a transseptal puncture would mitigate brain injury compared with a retrograde aortic approach. METHODS: The TRAVERSE trial (Transseptal Versus Retrograde Aortic Ventricular Entry to Reduce Systemic Emboli) was a multicenter randomized controlled comparative effectiveness trial. Patients with left ventricular arrhythmias undergoing catheter ablation procedures were randomly assigned to a transseptal puncture approach compared (1:1) with a retrograde aortic approach. The primary outcome was the presence of an acute brain lesion detected by magnetic resonance imaging. Secondary outcomes included clinically manifest complications, procedural efficacy, and 6-month neurocognitive assessments. RESULTS: Among the 62 patients randomly assigned to a retrograde aortic approach with postoperative brain magnetic resonance imaging, 28 (45%) exhibited an acute brain lesion compared with 19 of the 69 (28%) of those randomized to a transseptal puncture ( P =0.036). No differences in clinically manifest complications or procedural efficacy were observed. More patients in the retrograde aortic arm were categorized as having a high likelihood of cognitive impairment at 6 months (33% compared with 19% of those in the transseptal arm), but substantial loss to follow-up was present. CONCLUSIONS: Among patients undergoing left ventricular catheter ablation procedures, a transseptal approach reduced the risk of acute brain lesions by nearly half compared with a retrograde aortic approach without sacrificing safety or efficacy. Given a likely embolic pathogenesis, the brain magnetic resonance imaging findings may reflect a propensity to other organ damage; these findings may extend to other procedures requiring left ventricular entry. REGISTRATION: URL: https://www.clinicaltrials.gov ; Unique identifier: NCT03946072.

Migraine is one of the most prevalent and disabling neurological diseases, significantly affecting quality of life and productivity, as well as contributing to substantial societal costs. Recent innovations, including calcitonin gene-related peptide (CGRP) pathway inhibitors and onabotulinumtoxinA, have transformed migraine prevention by offering high efficacy and excellent tolerability, thus improving adherence. Clinical trials and real-world studies show that significant reductions in migraine frequency and, in some cases, complete migraine freedom is achievable. In this Position Statement, we advocate for raising the standards of migraine prevention by setting ambitious treatment goals aimed at optimal outcomes, such as migraine freedom or very low number of days with migraine or moderate/severe headache. We emphasize the importance of addressing residual migraine burden, highlighting that achieving a ≥50% reduction in monthly migraine days, although often considered a successful response, may not fully restore quality of life. Relying solely on percentage-based improvements can obscure the persisting impact of residual burden. This Position Statement does not want to change the standards for clinical trials but aims primarily at real-world clinical practice and proposes a shift from percentage-based measures of success to absolute goals while on treatment. We outline a framework that categorizes outcomes into four tiers: migraine freedom (no days with migraine or moderate-to-severe headache), optimal control (less than four days with migraine or moderate-to-severe headache), modest control (four to six days with migraine or moderate-to-severe headache) and insufficient control (more than days with migraine or moderate-to-severe headache). Focusing on residual burden while on treatment aims to further improve patient quality of life and drive innovation in preventive therapies and non-pharmacological approaches. By advocating for higher standards, this Position Statement, is not aimed primarily to drive reimbursement policies for migraine preventive treatments, but seeks to inspire clinicians, researchers and policymakers to prioritize ambitious goals in migraine prevention, ultimately enhancing patient outcomes and reducing the broader societal and economic impact of this debilitating condition.

Background and Objectives: Type 2 diabetes (T2D), closely associated with obesity, contributes to increased morbidity and mortality due to complications such as cardiometabolic disease. This review aims to evaluate the effectiveness of metabolic and bariatric surgeries (MBS) and endoscopic bariatric therapies (EBTs) in achieving diabetes remission and to examine key predictors influencing remission outcomes. Materials and Methods: This review synthesizes data from studies on MBS and EBT outcomes, focusing on predictors for diabetes remission such as preoperative insulin use, diabetes duration, HbA1c, and C-peptide levels. Additionally, predictive scoring systems, including the Individualized Metabolic Surgery (IMS), DiaRem, Advanced-DiaRem, ABCD, and Robert et al. scores, were analyzed for their utility in forecasting remission likelihood. Results: Key predictors of T2D remission include shorter diabetes duration, lower HbA1c, and higher C-peptide levels, while prolonged insulin use, and higher insulin doses are associated with lower remission rates. Scoring models like IMS and DiaRem demonstrate that lower scores correlate with a higher likelihood of remission, especially for procedures such as Roux-En-Y gastric bypass (RYGB). RYGB generally shows higher remission rates compared to sleeve gastrectomy (SG), particularly among patients with mild disease severity, while EBTs like ESG and IGBs contribute 5–20% total weight loss (TWL) and moderate glycemic control improvements. Conclusions: Both MBS and EBTs are effective for T2D management, with predictive scoring models aiding in individualized patient selection to optimize remission outcomes. Further research to validate these predictive tools across diverse populations could enhance treatment planning for both surgical and endoscopic interventions.

Backgrounds and Aims: HBV-related liver disease ranks as the seventh leading cause of mortality. Despite advances in prevention and treatment, global disparities in the burden of primary liver cancer (PLC) persist. We evaluate global trends in the prevalence, incidence, and death of HBV-related liver disease. Approach and Results: Data from the Global Burden of Disease Study 2021 evaluated acute HBV infection, HBV-related cirrhosis, and HBV-related liver cancer prevalence, incidence, and death. In 2021, there were 7.30 million cases of acute HBV, 283.64 million cases of HBV-related cirrhosis, and 288,110 cases of HBV-related PLC. HBV-related PLC accounted for 39% of the global incidence of PLC; Western Pacific had the highest rates for HBV-related PLC, with an incidence of 5.24 and a death rate of 4.38 per 100,000 population. Between 2000 and 2021, age-standardized incidence, prevalence, and death rates from HBV-related liver disease decreased; however, the incidence of HBV-related PLC rose in 65 countries from 2000 to 2021. In parallel, age-standardized prevalent rates from HBV-related PLC increased in Europe (annual percent change: 0.77%, 95% CI: 0.72%–0.81%) and the Americas (annual percent change: 1.05%, 95% CI: 1.00%–1.10%). Conclusions: From 2000 to 2021, decreases in HBV-related liver disease incidence and prevalence were observed; however, more than one-third of countries showed an increase in the incidence of HBV-related PLC. These findings highlight the need for strengthened HBV treatment efforts to reduce the risk of liver cancer.

Abstract Behavioral flexibility allows animals to adjust actions to changing environments. While the basal ganglia are critical for adaptation, the specific role of the external globus pallidus (GPe) is unclear. This study examined the contributions of two major GPe cell types—prototypic neurons projecting to the subthalamic nucleus (Proto GPe→STN neurons) and astrocytes—to behavioral flexibility. Using longitudinal operant conditioning with context reversals, we found that Proto GPe→STN neurons dynamically represent contextual information correlating with behavioral optimality. In contrast, GPe astrocytes exhibited gradual contextual encoding independent of performance. Deleting Proto GPe→STN neurons impaired adaptive responses to changing action-outcome contingencies without altering initial reward-seeking acquisition, highlighting their specific role in enabling behavioral flexibility. Furthermore, we discovered that Proto GPe→STN neurons integrate inhibitory striatal and excitatory subthalamic inputs, modulating downstream basal ganglia circuits to support flexible behavior. This research elucidates the complementary roles of Proto GPe→STN neurons and astrocytes in cellular mechanisms of flexible reward-seeking behavior.

Abstract The prevalence of steatotic liver disease (SLD) is rising globally, presenting a significant public health challenge. While non-contrast computed tomography (CT) scans hold promise for the opportunistic detection of SLD, their full potential has not been systematically explored. Clinical indicators, such as serum data, are crucial for monitoring the progression of SLD. We introduce a novel Multimodal AI for Opportunistic Steatotic Liver Disease Screening (MAOSS), which integrates demographic data, serum tests, conventional CT biomarkers, and deep learning-derived CT image features to enhance model performance and explainability. MAOSS employs an innovative learning strategy, trained on a large-scale dataset (n = 1,783) encompassing both histopathologically-confirmed SLD and radiologically-reported steatosis cases. The model underwent extensive retrospective validation using steatosis grades derived from both histology and MRI-PDFF across internal and external datasets (n = 913), as well as real-world patient data (n = 18,504). It accurately detected mild to severe steatosis (≥ S1) with AUCs ranging from 0.903 to 0.929. Notably, MAOSS significantly improved the overall AUC for radiologists (n = 11) from 0.709 to 0.798 (p < 0.001). Furthermore, MAOSS was applied to optimize the standard clinical pathway for a simulated primary risk screening of SLD patients (n = 1,192), resulting in the identification of 31% more patients at risk of steatohepatitis and liver fibrosis progression, highlighting its potential to enhance SLD detection and management.

Abstract Purpose Developing large‐scale, standardized radiographic registries for anterior cruciate ligament (ACL) injuries with artificial intelligence (AI) tools can enhance personalized orthopaedics. We propose deploying Artificial Intelligence for Knee Imaging Registration and Analysis (AKIRA), a trio of deep learning (DL) algorithms, to automatically classify and annotate radiographs. We hypothesize that algorithms can efficiently organize radiographs based on laterality, projection, identify implants and classify osteoarthritis (OA) grade. Methods A collection of 20,836 knee radiographs from all time points of treatment (mean orthopaedic follow‐up 70.7 months; interquartile range [IQR]: 6.8–172 months) were aggregated from 1628 ACL‐injured patients (median age 26 years [IQR: 19–42], 57% male). Three DL algorithms (EfficientNet, YOLO [You Only Look Once] and Residual Network) were employed. Radiograph laterality and projection (anterior‐posterior [AP], lateral, sunrise, posterior‐anterior, hip–knee–ankle and Camp‐Coventry intercondylar [notch]) were labelled by a DL model. Manually provided labels of metal fixation implants were used to develop a DL object detection algorithm. The degree of OA, both as measured by specific Kellgren–Lawrence (KL) grades, as well as based on a binarized label of OA (defined as KL Grade ≥2), on standing AP radiographs were classified using a DL algorithm. Individual model performances were evaluated on a subset of images prior to the deployment of AKIRA to registry construction using all ACL radiographs. Results The classification algorithms showed excellent performance in classifying radiographic laterality (F1 score: 0.962–0.975) and projection (F1 score: 0.941–1.0). The object detection algorithm achieved high precision–recall (area under the precision‐recall curve: 0.695–0.992) for identifying various metal fixations. The KL classifier reached concordances of 0.39–0.40, improving to 0.81–0.82 for binary OA labels. Sequential deployment of AKIRA following internal validation processed and labelled all 20,836 images with the appropriate views, implants, and the presence of OA within 88 min. Conclusion AKIRA effectively automated the classification and object detection in a large radiograph cohort of ACL injuries, creating an AI‐enabled radiographic registry with comprehensive details on laterality, projection, implants and OA. Study Design Cross‐sectional study. Level of Evidence Level IV.