Skip to main content

Weekly Respiratory Research Analysis

3 papers

This week’s respiratory literature highlights three cross-cutting advances: (1) a host-directed antiviral target (HGS) discovered by genome-wide CRISPRi with a repurposed compound demonstrating in vivo pan-coronavirus activity; (2) multimodal AI diagnostics that combine transcriptomic biomarkers (FABP4) with large language model EMR analysis to markedly improve lower respiratory tract infection (LRTI) diagnosis in critically ill patients; and (3) a pathomics whole-slide image score that predicts

Summary

This week’s respiratory literature highlights three cross-cutting advances: (1) a host-directed antiviral target (HGS) discovered by genome-wide CRISPRi with a repurposed compound demonstrating in vivo pan-coronavirus activity; (2) multimodal AI diagnostics that combine transcriptomic biomarkers (FABP4) with large language model EMR analysis to markedly improve lower respiratory tract infection (LRTI) diagnosis in critically ill patients; and (3) a pathomics whole-slide image score that predicts which lung squamous cell carcinoma patients derive survival benefit from first-line chemo‑immunotherapy. Together these studies accelerate host-targeted therapeutics, AI-enabled precision diagnostics, and image-derived treatment selection.

Selected Articles

1. Targeting the host factor HGS-viral membrane protein interaction in coronavirus infection.

85.5The Journal of Clinical Investigation · 2025PMID: 41401029

A genome-wide CRISPRi screen identified HGS as a conserved host factor that binds coronavirus M protein to enable ERGIC trafficking and virion assembly. M-derived peptides and the repurposed drug riboflavin tetrabutyrate (RTB) disrupted HGS–M binding, retained M in the ER, and blocked virion assembly, demonstrating broad pan‑coronavirus activity in vitro and in vivo.

Impact: Identifies a druggable, conserved host–virus interface (HGS–M) and provides a repurposed compound with in vivo efficacy, offering a host-directed antiviral approach with pandemic preparedness relevance.

Clinical Implications: If safety and PK/PD support translation, HGS inhibitors (peptides or RTB) could become broad‑spectrum antivirals less susceptible to viral mutation; next steps are ADME/toxicology and early-phase human trials.

Key Findings

  • Genome-wide CRISPRi screen implicates HGS as essential for pan-coronavirus infection.
  • HGS directly binds the viral M protein and facilitates ERGIC trafficking for virion assembly; HGS deficiency retains M in the ER and blocks assembly.
  • M-derived peptides and riboflavin tetrabutyrate (RTB) bind HGS, disrupt HGS–M interaction, and prevent virion assembly in vitro and in vivo.

2. Integrating a host biomarker with a large language model for diagnosis of lower respiratory tract infection.

84.5Nature Communications · 2025PMID: 41402257

In critically ill adults, combining the pulmonary transcriptomic biomarker FABP4 with GPT‑4 analysis of EMR free text produced a multimodal classifier with AUC ≈0.93 and accuracy 84% (96% in independent validation), outperforming biomarker-only, LLM-only, and clinician admission diagnosis. The approach improved discrimination between infectious and non-infectious respiratory failure.

Impact: Demonstrates a practical, externally validated multimodal diagnostic paradigm that leverages LLMs plus host biomarkers to meaningfully improve LRTI diagnosis in high‑risk ICU settings.

Clinical Implications: Implementation could accelerate accurate differentiation of infectious vs non-infectious respiratory deterioration, optimizing antimicrobial use and guiding targeted diagnostics; multicenter prospective impact studies are needed before deployment.

Key Findings

  • Combined FABP4 + GPT‑4 classifier achieved AUC 0.93±0.08 and 84% accuracy; validation cohort AUC 0.98±0.04 and 96% accuracy.
  • Combined model outperformed FABP4-only (AUC 0.84), LLM-only (AUC 0.83), and clinicians' admission diagnosis (accuracy 72%).
  • Independent validation reproduced performance, supporting reproducibility.

3. A pathomics model for predicting response to chemo-immunotherapy in lung squamous cell carcinoma: A multicenter study.

84.5Lung cancer (Amsterdam, Netherlands) · 2025PMID: 41421034

A pathomics model built from whole‑slide images and anchored to RNA‑seq predicted a T cell‑inflamed GEP and generated a pathomics score that identified LUSC patients who derive significant PFS and OS benefit from first‑line chemo‑immunotherapy versus chemotherapy. The score was validated in a prospective multicenter trial (AK105‑302) and replicated in independent cohorts, correlating with immune‑hot microenvironment features.

Impact: Operationalizes a scalable tissue‑based biomarker to guide first‑line chemo‑immunotherapy in LUSC without routine molecular assays, supported by prospective multicenter validation and biological concordance.

Clinical Implications: Integrating pathomics into pathology workflows could improve selection of LUSC patients for first‑line chemo‑immunotherapy, avoiding ineffective CIT in low‑score patients and optimizing outcomes; prospective implementation and cost‑effectiveness studies are next steps.

Key Findings

  • Pathomics score predicted T cell–inflamed GEP (AUC 0.80 training; 0.71 validation in TCGA LUSC).
  • Significant interaction between pathomics score and treatment in AK105‑302 for PFS (p=0.011) and OS (p<0.001); high‑score patients had large CIT benefit (PFS HR 0.31; OS HR 0.30 vs chemotherapy).
  • Findings replicated in two independent cohorts and associated with immune‑hot tumor microenvironment features.