Peachbot Bio™ is a branded Single Board Computer (SBC) platform and accompanying software stack developed to explore decentralized execution of graph-based machine learning models for biomedical network analysis. The platform is designed to support local training and inference of Graph Neural Networks (GNNs) on molecular interaction graphs, reducing dependence on centralized cloud infrastructure.
Peachbot Bio™ is intended as a research and systems-validation platform, aligned with the accompanying academic study on edge-based graph neural network feasibility.
The Peachbot Bio™ platform provides a reference implementation for executing GNNs over a protein–protein interaction (PPI) network consisting of:
- 1,603 genes (nodes)
- 2,757 molecular interactions (edges)
The repository demonstrates how TCGA-scale transcriptomic datasets can be mapped onto a graph representation and processed locally on SBC-class hardware. The focus is on numerical stability, convergence behavior, memory constraints, and inference latency, rather than clinical deployment.
-
On-Device Graph Neural Network Execution Supports local inference and constrained training of GNNs with >1,600 nodes on Peachbot Bio™ hardware.
-
Edge-Based Bioinformatics Workflows Enables local processing of publicly available cancer genomics datasets (e.g., TCGA) without external data transfer.
-
Low-Latency, Bounded Inference Provides consistent inference latency suitable for edge computing environments.
-
Data Locality by Design All computation is performed locally on the SBC, supporting data governance and experimental privacy requirements.
Validation experiments on Peachbot Bio™ were conducted in alignment with the accompanying paper and are explicitly divided into two categories:
-
Purpose: Evaluate numerical stability, convergence behavior, and memory utilization under controlled conditions.
-
Method: Synthetic amplification of selected node features to create a high signal-to-noise regime.
-
Interpretation: These experiments validate hardware and software feasibility only and are not clinically meaningful.
-
Dataset Example: TCGA-BRCA (1,502 samples with complete annotations)
-
Metrics Reported: Balanced accuracy, ROC-AUC, precision (consistent with published network-based oncology studies)
No claims of clinical diagnostic accuracy or medical decision-making are made.
Ensure the Peachbot Bio™ SBC is flashed with the supported Linux-based operating system and that the Python environment for graph neural network execution is active.
Run the edge-based GNN pipeline locally:
python run_gnn_pipeline.py --device local_gpu --mode biologicalSynthetic systems validation experiments can be executed using:
python run_gnn_pipeline.py --device local_gpu --mode synthetic-
Local Data Ingestion Loads transcriptomic and clinical metadata from local storage.
-
Graph Construction Converts patient-level data into graph objects aligned with the PPI topology.
-
Graph Convolution Performs sparse, normalized message passing optimized for SBC-class hardware.
-
Result Export Outputs predictions and evaluation metrics as local CSV files.
Peachbot Bio™ is intended for research, systems evaluation, and educational purposes only.
- ❌ Not a medical device
- ❌ Not validated for clinical diagnosis
- ❌ Not approved for patient care
All outputs should be interpreted as computational research results, not clinical recommendations.
- Platform Name: Peachbot Bio™
- Hardware: Peachbot Bio™ SBC (v1.0)
- Lead Developer: Swapin Vidya
- Research Context: Independent academic and systems research
This project description is fully aligned with the paper:
Decentralized Network Intelligence: Evaluating the Feasibility of Graph Neural Networks on Edge Hardware for Clinical Oncology
No claims in this repository exceed or contradict the scope, results, or limitations stated in the manuscript.