A Computational Workflow to Identify Small Molecules That Mimic Exercise-Associated Gene Expression Signatures Using MoTrPAC and LINCS

Abstract

Exercise produces coordinated, tissue-dependent changes in gene expression, but it remains unclear which small molecules reproduce those transcriptional responses. This study protocol describes a computational workflow to identify candidate compounds whose perturbational gene-expression signatures mimic exercise-associated signatures measured by the Molecular Transducers of Physical Activity Consortium (MoTrPAC). Differentially expressed genes from a prespecified MoTrPAC comparison will be separated into upregulated and downregulated sets, analyzed for pathway enrichment, and submitted to SigCom LINCS for comparison with perturbational signatures from the Library of Integrated Network-Based Cellular Signatures (LINCS) L1000 resource. Candidate mimickers will be prioritized using the direction and strength of signature concordance, consistency across relevant conditions, and compound metadata. Public chemical and drug resources will then be queried to annotate availability and identifiers. The workflow will generate enrichment tables, heatmaps, ranked signature-search tables, a bar plot of leading candidates, and a compound-availability table. The analysis is designed to distinguish transcriptional mimicry from reversal of an exercise signature and from demonstrated physiological benefit. No computational results are reported here because the underlying MoTrPAC differential-expression results and SigCom LINCS search outputs have not yet been analyzed in this manuscript. This protocol provides a reproducible framework for generating hypotheses about exercise-associated molecular perturbagens; experimental validation will be required before any candidate can be described as a functional exercise substitute.

Introduction

Exercise elicits molecular responses across tissues and biological systems. The Molecular Transducers of Physical Activity Consortium (MoTrPAC) was established to characterize these responses across exercise modalities, time points, and molecular layers, creating a resource for investigating the mechanisms associated with physical activity.¹ MoTrPAC publications and data resources provide a foundation for selecting well-defined exercise-associated gene-expression contrasts.[^2,^3]

Perturbational transcriptomics offers a complementary strategy: instead of asking only which genes change after exercise, one can ask which experimental perturbations produce a similar pattern. LINCS L1000 profiles gene-expression responses to small molecules and other perturbations, while SigCom LINCS provides search functionality for comparing an input signature with reference signatures.² A compound whose signature resembles an exercise-associated signature may help identify molecular pathways or candidate interventions for further study.

However, signature similarity is not equivalent to a biological or clinical substitute for exercise. Transcriptomic responses depend on cell type, dose, exposure duration, assay, and experimental context. Moreover, a compound can mimic one subset of an exercise response while opposing another. Accordingly, candidate ranking must preserve the distinction between mimickers and reversers, account for tissue and time-point context, and present uncertainty rather than overstate therapeutic implications.

The objective of this work is to specify a transparent workflow that integrates MoTrPAC differential-expression results with SigCom LINCS perturbational signatures and public compound resources. The primary output will be a ranked list of candidate small molecules with evidence of transcriptional concordance, accompanied by pathway-level interpretation and compound annotations.

Research questions

1. Which LINCS perturbations most closely mimic the direction of gene-expression changes observed in a selected MoTrPAC exercise comparison?
2. Which biological pathways are enriched among exercise-associated upregulated and downregulated genes?
3. Are candidate mimickers supported across more than one relevant signature or experimental context, or are they specific to a single cell line, dose, or time point?
4. Which prioritized compounds can be mapped to public chemical or drug records, and what identifiers and availability information can be verified?
5. How sensitive are the candidate rankings to the choice of tissue, exercise modality, sampling time, gene-selection threshold, and signature-search parameters?

Methods

Study design and analysis scope

This is a computational secondary-analysis protocol. The analysis will use publicly accessible MoTrPAC gene-expression results and perturbational signatures indexed by SigCom LINCS. Before analysis, the target MoTrPAC contrast must be documented with its species, tissue, exercise modality, intervention type (acute bout or training), sampling time, comparison group, and source dataset or publication. These attributes are essential because an exercise response is not a single universal signature.

The primary analysis will use one explicitly selected contrast. Additional contrasts may be analyzed as sensitivity analyses, but their results will be kept separate unless a prespecified method justifies integration. The source dataset's data-use terms will be checked before downloading or analyzing data, and only data permitted for the intended use will be included.

MoTrPAC differential-expression signature

Differential-expression results will be obtained from the relevant MoTrPAC data release or a peer-reviewed analysis that provides the necessary gene-level statistics. The source, release/version, gene identifier type, contrast definition, statistical model, covariates, and significance criteria will be recorded. Where possible, effect sizes and adjusted P values will be retained rather than reducing the input immediately to a binary gene list.

Genes will be mapped to the identifier system accepted by the enrichment and signature-search services. Mapping will be documented, duplicate mappings handled consistently, and unmapped genes reported. The primary directional gene sets will comprise genes increased after exercise and genes decreased after exercise, using a threshold selected before inspecting compound-search results. If a ranked gene-level statistic is available, a rank-based sensitivity analysis will be considered alongside thresholded sets.

Pathway enrichment

Upregulated and downregulated genes will be analyzed separately for pathway or gene-set enrichment. The gene-set library, release/version, background universe, multiple-testing correction, and significance threshold will be recorded. Where the method supports it, the background will be restricted to genes measurable or testable in the source experiment rather than assuming all annotated genes were eligible for detection.

Results will be reported in tables containing the pathway name, identifier, direction, overlap size, enrichment statistic, raw P value, adjusted P value, and contributing genes, as supported by the selected tool. Heatmaps will summarize selected pathways across the analyzed contrasts, with the plotted quantity and scaling explicitly labeled. Enrichment will be interpreted as statistical overrepresentation, not proof that a pathway is activated or causally responsible for the physiological response.

SigCom LINCS signature search

The exercise signature will be submitted to SigCom LINCS using its signature-search interface or documented API.² The search direction will be set to identify mimickers: perturbational signatures in which genes increased by exercise tend to be increased and genes decreased by exercise tend to be decreased. Reversers, which show the opposite directional pattern, will be treated as a distinct result category and will not be mislabeled as mimickers.

For each returned signature, the analysis will capture the perturbagen name and identifier, perturbation type, cell line or biological context, dose, exposure time, signature identifier, search score or z-scores, and any available quality or metadata fields. The exact search request, date, service/API version where available, and downloaded result files will be retained for reproducibility. Because the reference signatures may be derived from cultured cell lines rather than the tissue in which exercise was measured, context mismatch will be explicitly considered during interpretation.

The primary candidate table will rank results using the service's documented signature-search score and direction labels. If multiple signatures correspond to the same compound, they will remain traceable to their underlying experimental conditions. Any compound-level aggregation will be prespecified and will not simply select the most favorable result without reporting the underlying variability. A bar plot will show the leading candidate signatures or compounds and the exact score used for ranking. The number of candidates displayed and any filtering rules will be stated in the figure legend.

Candidate prioritization and sensitivity analyses

Candidate prioritization will consider (1) directional concordance with the exercise signature, (2) score strength and statistical evidence returned by the search service, (3) reproducibility across available signatures or relevant conditions, (4) interpretability of perturbation metadata, and (5) successful mapping to a public compound record. These criteria will be reported separately where possible rather than combined into an opaque score.

Sensitivity analyses will examine whether the leading candidates persist when reasonable gene-selection thresholds or alternative MoTrPAC contrasts are used. If the source provides several time points, modality-specific and time-specific signatures will be compared rather than collapsed prematurely. Candidate stability will be described as robustness of computational ranking, not as independent experimental replication.

Compound annotation and availability

Prioritized perturbagens will be mapped to public chemical or drug resources using exact names and, where available, stable identifiers such as PubChem Compound IDs, DrugBank identifiers, or other database accessions. Synonyms and salt forms will be resolved cautiously; ambiguous matches will be marked rather than silently merged. A compound-availability table will include the perturbagen name, normalized name, identifier, database source, record URL, match confidence, and verified availability information.

“Available” will be defined operationally according to the resource queried—for example, a public database record, catalog listing, or documented source of material. A database record alone does not establish current commercial availability, regulatory approval, safety, suitable dosing, or suitability for human use. These claims will be made only when directly supported by an authoritative, current source.

Reproducibility and reporting

The final report will include source access dates, dataset and software versions, analysis parameters, identifier-mapping rules, query inputs, and the raw or processed result tables needed to reproduce the figures. The planned outputs are: (1) a table of the selected MoTrPAC contrast and signature genes; (2) up- and downregulated gene-set enrichment tables; (3) pathway heatmaps; (4) a ranked SigCom LINCS result table; (5) a bar plot of leading mimicking signatures or compounds; and (6) a compound-annotation and availability table. No numerical values, candidate rankings, or visual results will be reported unless they are produced from actual analysis outputs.

Results status

This manuscript describes the analysis framework and does not report completed computational findings. The underlying MoTrPAC contrast, enrichment outputs, SigCom LINCS results, and verified compound annotations have not been supplied or retrieved for this draft. Consequently, no claim is made that a particular compound mimics exercise, that any pathway is significantly enriched in a specific contrast, or that any candidate is available. The Results section should be completed only after the workflow has been executed and its outputs checked.

Discussion

Integrating exercise-associated transcriptional changes with perturbational signatures can prioritize small molecules for mechanistic follow-up. The central strength of the proposed workflow is its explicit separation of directional mimicry from reversal, combined with pathway-level interpretation and traceable compound annotation. A ranked list can support hypothesis generation and help identify perturbations for follow-up experiments.

Several limitations are important. First, exercise responses vary by tissue, modality, time, species, training status, and study design. A signature selected from one context should not be treated as a universal exercise response. Second, LINCS profiles are generated under defined experimental conditions, often in cell lines that do not reproduce the tissue environment or systemic physiology of exercise. Third, the L1000 platform directly measures a landmark-gene subset and infers additional expression values; results therefore inherit platform and inference limitations.² Fourth, signature concordance does not establish that a compound reproduces the downstream physiological effects of exercise, including cardiovascular, metabolic, musculoskeletal, and behavioral adaptations. Finally, database annotations and catalog listings can change and should be verified at the time of reporting.

The next stage is to execute the analysis using a prespecified MoTrPAC contrast, save the service responses and enrichment outputs, and complete the result tables and figures. High-priority candidates should then be evaluated in an appropriate experimental system with matched dose and exposure conditions, independent transcriptional measurements, and phenotypic assays relevant to the proposed mechanism. Safety, pharmacokinetics, tissue exposure, and potential off-target effects must be evaluated before translational claims are considered.

The intended interpretation is therefore narrow: this workflow identifies candidate perturbations that may reproduce selected aspects of an exercise-associated gene-expression signature. It does not establish that a compound substitutes for physical activity or delivers equivalent health benefits.

Conclusion

A reproducible integration of MoTrPAC exercise signatures, pathway enrichment, SigCom LINCS searches, and public compound annotation can generate a prioritized set of hypotheses about small molecules associated with exercise-like transcriptional responses. Completion of the empirical analysis and experimental validation are required before reporting specific candidates or drawing biological or therapeutic conclusions.

Data and code availability

No new dataset or analysis output is included in this protocol draft. The source data should be obtained from the relevant MoTrPAC Data Hub release and used in accordance with its access and data-use terms. SigCom LINCS provides public signature-search functionality and documented data access. Analysis scripts, exact query parameters, and generated tables should be deposited in a stable public repository when the empirical analysis is complete.