Exersomes

Exersomes is a comprehensive Go-based bioinformatics and simulation framework designed to model, analyze, and predict the molecular responses to exercise.

The project focuses on exerkines—signaling molecules (proteins, metabolites, miRNAs, RNAs) released by various tissues (like skeletal muscle, mitochondria, and placenta) during physical exertion. By integrating real-world biological data with programmatic simulations and a powerful Protein-Nucleic acid Language Model (PNLM), Exersomes aims to map the complex molecular networks of exercise physiology for personalized precision medicine, gene therapy, and disease modeling.

🧬 Key Features

• Molecular Type Modeling: Strongly-typed Go structs representing diverse exercise-responsive exerkines molecules including Proteins (e.g., IL-6, IL-10, PPARG), Metabolites (e.g., 12,13-diHOME, Lac-Phe, Lactate, AICAR, BAIBA, 2-OHB, 3-OHB), miRNAs, RNAs (e.g., exerkines), and Vesicles (e.g., exersomes, cargo (RNAs, proteins).
• Protein-Nucleic Language Model (PNLM): Natively integrates with transformer-based architectures to measure and embed RNA trajectory velocities, protein expression levels, and lipid/metabolite cross-talk for high-precision exerkine generation.
• Temporal Dynamics & Circulating Factors: Calculates physiological half-lives, acute vs. chronic release times, and tracks bloodstream factors (e.g., Epinephrine, Cortisol, Lactate, BDNF) during and post-exercise.
• Mitochondrial Signaling (Mitokines): Simulates the release and downstream effects of mitokines (e.g., PGC-1α, TFAM, FGF21, GDF15, mtROS) in response to bioenergetic demands and mitochondrial stress.
• Exercise Prescription Simulation: Models specific workout protocols encompassing intensity and duration to predict quantitative molecular responses over time.
• Automated NCBI Integration: Uses NCBI E-utilities to dynamically fetch gene references, protein accessions, FASTA sequences, KEGG/BioSystems pathway maps, and PubMed functional insights for targeted exerkines.
• Translational Consortium Interoperability: Capable of integrating structural and spatial molecular responses for mapping to MoTrPAC, the HuBMAP Human Reference Atlas, and HTAN exercise immuno-oncology models.

🛠 Prerequisites

• Go 1.24.1 or higher
• NCBI E-utilities (Entrez Direct): Required for the Go-based data fetching pipeline (esearch, efetch, elink).
  • macOS / Linux Installation: Run the official setup script in your terminal:

    sh -c "$(curl -fsSL https://ftp.ncbi.nlm.nih.gov/entrez/entrezdirect/install-edirect.sh)"

  • Note: After installation, follow the terminal prompt to add the edirect folder to your system's PATH. Your terminal needs to be able to find these commands globally for the Go application to work.

🚀 Installation & Setup

1. Clone the repository and navigate to the project directory.
2. Run the project directly without building a permanent executable:

   go run ./exersomes/exersomes.go

💻 Usage

1. Retrieving RNA & Protein Sequences (NCBI Pipeline)

To automatically fetch the FASTA sequences and biological data for your exerkines, ligands, and receptors:

1. Edit the Input File: Open data/raw_data/exerkines.txt and ensure your target genes are listed (one per line). You can include exerkines (e.g., IL6), ligands (e.g., IGF1), and their corresponding receptors (e.g., IL6R, IGF1R).
2. Run the Pipeline: Execute the main Go program from your terminal:

   go run ./exersomes/exersomes.go

3. Collect Outputs: The program uses parallel workers to query NCBI databases and will generate several files in your root directory:
   • rna_sequences.fasta: Ready-to-use human and rat mRNA nucleotide sequences.
   • protein_sequences.fasta: Amino acid sequences for structural modeling.
   • protein_info.tsv: Protein accessions and molecular weights.
   • rna_references.tsv: RNA RefSeq accessions and descriptions for human and rat.
   • pathway_maps.tsv & functional_insights.tsv: Mined functional insights from PubMed and GO.

2. Physiological Simulations

Programmatically integrate the additional omes, cardiometabolism, placental components, and organokiome to infer and explain molecular responses:

• Adipomyokinome: Use InferAdipomyokines(exerciseType, intensityPercent, durationMinutes, timePointsCount) to map the rise and decay of adipomyokines like IL-6 in obesity and cancer.
• Mitokine Signalling: Use CalculateMitokineResponse to estimate the change in mitokine levels based on exercise parameters and training status for healthy pregnancy and pregnancy complications (gestational diabetes and preclampsia).
• Organokinome: Use CalculateOrganokineResponse to estimate the change in organokine levels based on exercise parameters, gut microbiota, and training status.

3. Language Model (LM) Integrations

Exersomes provides a unified Go client interface to orchestrate state-of-the-art machine learning models for deep molecular analysis:

• PNMLM (Protein-Nucleic Acid-Metabolite Language Model): Generate multi-modal latent embeddings combining RNA velocity, protein expression, and localized lipids. Predict systemic exersome assembly and decode personalized exercise prescriptions.
• ESM-2 (Evolutionary Scale Modeling): Perform zero-shot structural prediction, inter-residue contact mapping, and generate novel viable protein sequences (exerkines/ligands).
• EDM (Equivariant Diffusion Model): Generate highly precise 3D biochemical structures (metabolites, lipids) natively adhering to organic chemoinformatics constraints (C, H, O, N, S).

🗂 Project Structure

exersomes/
├── exersomes.go                             # Main NCBI data fetching entrypoint
├── molecular_types/                         # Core exerkine models (miRNA, RNA, Proteins)
├── organs/
│   ├── muscle/                              # Skeletal muscle
|   ├── adipose/                             # Adipose tissue 
│   ├── cardiovascular/bloodstream/          # Circulating blood factors & temporal decay
│   └── placenta/                            # Mitokines and mitochondrial signaling
├── models/                                  # Deep learning architecture and clients
│   ├── lm.go                                # PNMLM, ESM-2, and EDM language model clients
│   ├── latentspace.go                       # VAE latent space representations
│   └── encoder.go                           # Exersome multi-omic cargo encoder
└── src/                                     # APIs, data preprocessing, and network analysis

📄 License

Open-source. Please ensure compliance with NCBI's data usage guidelines when running the automated E-utilities fetching pipeline.

If you use Exersomes in your research, please cite:

Gomez, DJ. et al. (2026). Exersomes: A computational tool for analyzing exercise-induced signaling molecules. [Software]. Available from https://github.com/djgomezsantos/Exersomes