Short-term molecular assessment of peripheral blood transcriptomics and urinary metabolomics using machine learning and Bayesian network modeling in a randomized controlled study of smokers after switching to heated tobacco products

Published Jun 16, 2026 on Dryad. https://doi.org/10.5061/dryad.3j9kd520x

Data files

Jun 16, 2026 version files 155.83 KB

Abstract

Background: Omics approaches can provide mechanistic insights into biological changes after changes in smoking. However, data on short-term molecular changes after changes in tobacco product use are limited.

Objectives: To compare 5-day molecular changes in peripheral blood gene expression and urinary metabolites among participants who continued smoking combustible cigarettes (CS), switched to heated tobacco products (HTPs; DT3.0a or THS), or stopped smoking (SS) and to explore probabilistic relationships among selected molecular features using machine learning and Bayesian networks.

Design: We performed a secondary analysis of stored specimens collected from a randomized, controlled, impatient, parallel-group intervention trial in which participants underwent a 5-day intervention.

Methods: Peripheral blood transcriptomics (mRNA and miRNA) were measured at baseline (Day 1) and at the end of intervention (Day 6). Urinary metabolomics were measured on Day 6. Differential expression screening used exploratory thresholds. Machine learning was used for feature selection and classification of CS vs SS signatures; Bayesian network models represented conditional probabilistic dependencies among molecular features and downstream events under explicit modeling assumptions.

Results: Compared with the CS groups, both HTP groups and the SS group showed shifts in blood transcriptomic signatures and urinary metabolites consistent with reduced oxidative stress and inflammatory signaling during the 5-day intervention. Classification models tended to place HTP profiles closer to those of the SS group than the CS group for most modeled processes, while plaque-related outputs remained less conclusive.

Conclusion: Over 5 days, switching from smoking to HTPs was associated with molecular changes in peripheral blood gene expression and urinary metabolites that partially resembled those observed with smoking cessation. These exploratory findings do not establish long-term physiological status and should be interpreted within the measured matrices and the study’s limited duration and sample size. Further studies to confirm these findings in an external cohort are required.

Dataset DOI: 10.5061/dryad.3j9kd520x

Description of the data and file structure

This dataset is derived from stored urine samples collected in a randomized, controlled, parallel-group inpatient study with a 5-day intervention period involving adult smokers. Participants were randomized to continue smoking, switch to heated tobacco products, or abstain from smoking, and biospecimens were collected at baseline (Day 1) and at the end of the intervention (Day 6).

Files and variables

File: HRS-2017_Eicosanoids.xlsx

Description:
This file contains sample-level eicosanoid measurements. The file includes subject identifiers and study day, followed by analyte-specific variables for measured concentration, comment flags, and assay thresholds (LOD, LLOQ, and ULOQ). The file also contains creatinine-related variables.

Variables
Subject ID: Unique subject identifier.
Day: Study day at sample collection. Values shown in the file include D1 and D6.

Creatinine (mg/L)_Concentration: Measured creatinine concentration (mg/L).
Creatinine (mg/L)_Comment: Comment flag for the creatinine measurement. The header indicates the possible flags are <LOD, <LLOQ, and >ULOQ.
Creatinine (mg/L)_LOD: Limit of detection for creatinine (mg/L).
Creatinine (mg/L)_LLOQ: Lower limit of quantification for creatinine (mg/L).
Creatinine (mg/L)_ULOQ: Upper limit of quantification for creatinine (mg/L).

For each eicosanoid analyte (e.g., 11-dh-TXB2, 2,3-d-TXB2, 8-i-PGF2a, PGEM, PGDM, PGF2a, 2,3-d-8-i-PGF2a, LTE4, and HETE), the following variable pattern is used:

[Analyte name] (ng/mL)_Concentration: Measured analyte concentration in ng/mL.
[Analyte name] (ng/mL)_Comment: Comment flag for the analyte measurement. Possible flags are <LOD, <LLOQ, and >ULOQ.
[Analyte name] (ng/mL)_LOD: Limit of detection for the analyte (ng/mL).
[Analyte name] (ng/mL)_LLOQ: Lower limit of quantification for the analyte (ng/mL).
[Analyte name] (ng/mL)_ULOQ: Upper limit of quantification for the analyte (ng/mL).

Missing values / special values:
Blank cells indicate missing values where applicable. The Comment columns explicitly indicate censored values such as <LOD, <LLOQ, and >ULOQ.

File: HRS-2017_Metabolomics.csv

Description:
This file contains sample-level metabolomics data. The file includes group, subject identifier, study day, and measurement unit, followed by metabolite variables.

Variables
Group: Study group. Group values shown in the data include CC and DT. In the associated study documentation, the study groups are defined as DT3.0a (switched to DT3.0a), THS (switched to THS), SS (stopped smoking), CC (continued to smoke own brand of cigarettes), mDT3.0a (switched to a menthol DT3.0a), and mCC (continued to smoke own brand of menthol cigarettes).
Subject ID: Unique subject identifier.
Day: Study day at sample collection. Values shown in the file include D6.
unit: Unit of measurement recorded in the file. Values shown include µmol/mmol Cr.

Metabolite variables:
Each remaining column corresponds to one metabolite. The variable name is the metabolite name, and the recorded value is the measured level in the unit given in the unit column. Examples include 1,3-Diaminopropane, 1-Methyl-2-pyrrolidone, 1-Methyladenosine, 1-Methylnicotinamide, Betaine, and Betaine aldehyde_H2O.

Abbreviations used in metabolite names:
Cr. in the unit column refers to creatinine, as shown in the same data package where creatinine is explicitly labeled as Creatinine.
H2O appears in some metabolite names exactly as part of the variable name (e.g., Betaine aldehyde_H2O).
divalent appears in some variable names exactly as part of the metabolite name (e.g., Acetyl CoA_divalent, CoA_divalent, Anserine_divalent).

Missing values / special values:
Blank cells indicate missing values where applicable. Blank entries are visible in the file.

General note:
Variable names were imported automatically from the uploaded files. In the metabolomics file, each metabolite column stores a measured value in the unit specified in the unit column. In the eicosanoids file, each analyte is represented by concentration, comment, LOD, LLOQ, and ULOQ fields.

Code/software

No custom code is included in the current data package; the deposited materials consist of tabular data files in CSV and XLSX formats.

Access information

Other publicly accessible locations of the data:

Related transcriptomic datasets from the same study have been deposited in the NCBI Gene Expression Omnibus under accession numbers GSE328687 and GSE329249.

Data was derived from the following sources:

This dataset is part of the same study that generated the mRNA-seq and miRNA-seq datasets deposited in the NCBI Gene Expression Omnibus under accession numbers GSE328687 and GSE329249.

Human subjects data

Participants were identified using study identification numbers, and direct identifiers (e.g., names and medical record numbers) were removed from source document copies.