Comments
Write a comment-
Comment by Emerald Bee
- Published
- License
- CC BY 4.0
Updated review:
This study leverages a multi-omics approach, combining host transcriptomic profiling and airway microbial analysis, to identify early diagnostic biomarkers of pediatric lower respiratory tract infections (LRTI). Using publicly available data from a cohort of 261 critically ill children, Wu et al. identify key pathogens such as RSV and Haemophilus influenzae and develop a streamlined seven-gene host signature that performs comparably to larger gene panels. These findings are validated in an independent prospective cohort, the RASCALS study, demonstrating promising diagnostic accuracies. The integration of machine learning and network analyses offers deep insights into host immune pathways and host-microbe interactions, and the proposed gene signature has potential clinical utility.
However, many of the analyses substantially overlap with prior studies using the same public data set, and several findings are presented as novel despite having been previously reported. Additionally, aspects of the analytical methodology are insufficiently described or appear inconsistent with standard practices in the field. Major revisions are required to clarify methodological choices, accurately contextualize prior work, and clearly delineate the study’s novel contributions.
Major Comments:
● The authors mischaracterize findings from Mick et al. (2023) one of the papers with originally introduced the publicly available dataset used in this study.
○ In section 3.4 “Key Microbial Strains Associated with LRTI Progression” the authors state that Mick et al. reported LRTI progression or outcomes; however, this is factually incorrect
○ Authors performed a differential abundance analysis using methods that are not standard in the field and reported that RSV and H. influenza were more common in patients with LRTI vs those without.
○ This is the same finding described by Tsitsiklis et al 2022 but this overlap is not acknowledged
○ In the “Research in Context” section, authors state that "Mick et al. showed that a 14-gene host signature could separate bacterial from viral infections but did not address host–microbe dynamics." However, Mick et al’s main point was to integrate host and microbial data to improve all-cause LRTI diagnosis, and the 14-gene signature was not designed to distinguish bacterial from viral etiologies
Authors should accurately describe prior work and clearly articulate how their analyses and conclusions extend beyond existing studies.
● In the methods section, the authors state that samples were taken within 24 hours of patient enrollment, but do not specify the timing of sample collection relative to disease onset, which would clarify the diagnostic window of these biomarkers.
○ Authors should include more details in the methods section about when samples were collected relative to symptom onset, hospital admission, or initiation of therapy, as these factors influence biomarker performance.
● Authors do not include an analysis or discussion on potential confounding factors such as prior antibiotic treatment, co-infections, or underlying comorbidities. This could involve subgroup analyses if data are available.
○ Authors should add a subsection in the discussion discussing the influence of these factors on biomarker performance or outlining strategies for future research to control or adjust for these variables.
● The authors mention microbial detection and associated analyses in section 3.4, where they discuss the identification of dominant microbial strains such as RSV and Haemophilus influenzae. Specifically, they note that viral counts were adjusted against water controls and that microbial counts were filtered accordingly: "To control for contaminants, microbial counts were adjusted against water controls. After filtering, viruses were detected in 91% of LRTI samples (107/117) versus 16% of controls (8/50) ..." However, the detailed methodology for adjusting microbial load is not elaborated on.
○ Authors could clarify by specifying the exact methods used for contamination control, such as the thresholds set or the statistical models employed during microbial count adjustment, and how microbial abundance was quantified and normalized post-filtering. A supplementary figure showing the pre- and post-filtering microbial distributions could be included to improve clarity.
● The authors report identifying ~144-180 differentially expressed genes between definite LRTI and no-LRTI cases. This is markedly lower than results reported in prior analyses of the same dataset, which identified several thousand differentially expressed genes using conservative methods such as limma-voom with Benjamini–Hochberg correction. This discrepancy raises concerns about model specification, filtering criteria, or multiple testing correction and should be explicitly addressed
Minor Comments:
● The integration of multi-omics data and classification algorithms is a notable strength of this study, and the development of a simplified seven-gene panel is highly relevant for clinical translation. However, future studies should focus on prospective validation in more diverse and larger cohorts, including different age groups and healthcare settings.
○ Authors could explicitly mention plans for such validation or discuss ongoing efforts to test the biomarkers’ performance elsewhere.
● The focus on key pathogens, such as RSV and H. influenzae, enhances clinical relevance; however, expanding pathogen detection to encompass other atypical or emerging pathogens would improve comprehensive diagnostic capacity.
○ The authors could add a paragraph in the discussion to address the potential of expanding their pathogen panel or mention ongoing or planned studies aimed at including wider pathogen detection.
Competing interests
The author of this comment declares that they have no competing interests.