298. Host Response Profiling from Clinical Metagenomic Sequencing Data for Diagnosis of Central Nervous System Infections

298. 基于临床宏基因组测序数据的宿主反应分析在中枢神经系统感染诊断中的应用

阅读:2

Abstract

BACKGROUND: Clinical metagenomic next-generation sequencing (mNGS) testing of cerebrospinal fluid (CSF) increases diagnostic yield for suspected central nervous system (CNS) infections. Nevertheless, up to 45% of cases remain unknown despite extensive testing and 6 months of follow-up. We developed artificial intelligence-machine learning (AI-ML) classification models (classifiers) based on RNA gene expression / host response data from CSF mNGS testing to enhance diagnostic performance. [Figure: see text] METHODS: From June 2016 and April 2023, 464 CSF mNGS results from UCSF patients with a confirmed viral (n=175), bacterial (n=91), fungal (n=44), or non-infectious (n=155) diagnosis were randomly divided into training and test subsets in an 80:20 ratio (Figure 1A). Clinicians adjudicated their confidence in the final diagnosis based on blinded medical chart review and laboratory test results. Gene (feature) selection was carried out using high-confidence samples, followed by 50-fold cross-validation. All possible pairwise comparisons were performed to generate sub-classifiers which were then integrated into a consensus classifier. Performance metrics were obtained by running the consensus classifier on the independent test subset. A separate classifier was developed for parasitic infections (n=24) using a “leave-one-out” algorithm to assess performance (Figure 1B). Classifier results were displayed as a score ranging from 0 to 10, corresponding to specificities of ≤70% to ≥99%. [Figure: see text] RESULTS: Classifier accuracy based on the test set was 83%, with individual area under the curve (AUC) scores ranging from 0.88-0.93 for each category (Figure 1C). The function of the selected genes corresponded well with the category (Figure 2). Classifier examples include patients with (i) culture-negative CNS tuberculosis, (ii) persistent Toxoplasma gondii  infection despite 19 days of treatment, (iii) a rare autoimmune syndrome, and (iv) a chronic enterovirus infection misclassified as an atypical bacterial infection, suggesting that acute and chronic host response profiles differ (Figure 3). [Figure: see text] CONCLUSION: Host response profiling with AI-ML classifiers complements mNGS testing and can enhance diagnostic yield for unexplained CNS syndromes. DISCLOSURES: Charles Chiu, MD, PhD, Abbott Laboratories, Inc: Grant/Research Support|Biomeme: Advisor/Consultant|Biomeme: Board Member|BiomeSense: Advisor/Consultant|BiomeSense: Board Member|Delve Bio: Advisor/Consultant|Delve Bio: Board Member|Delve Bio: Grant/Research Support|Flightpath Biosciences: Advisor/Consultant|Flightpath Biosciences: Board Member|Mammoth Biosciences: Advisor/Consultant|Mammoth Biosciences: Board Member|Pathogen detection using next generation sequencing: US patent 11380421|Poppy Health: Advisor/Consultant|Poppy Health: Board Member

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。