RDP++ (Recombinant Detection Prediction)

A C++ application for recombinant detection prediction using neural networks with ONNX Runtime.

Overview

RDP++ is a tool that uses a trained neural network model to predict recombinant classes based on a set of input features. The system consists of two main components:

1. PSNN (Prediction System Neural Network): Processes standardized input data and runs inference using an ONNX model.
2. Tester: Prepares test data and interacts with PSNN to generate predictions.

The system reads feature data from a shared file, processes it, and outputs classification probabilities for different recombinant classes.

Requirements

• C++17 compatible compiler (g++, MSVC, etc.)
• ONNX Runtime (version 1.21.1 or compatible)
• For GPU acceleration: CUDA compatible GPU with appropriate drivers

Project Structure

• PSNN.cpp and PSNN.h: Main prediction system that processes input data and runs the neural network model
• tester.cpp: Tool for generating test data and running the prediction system
• RDP_TripleNN.onnx: The trained neural network model
• sharedData.txt: Input data file shared between components
• prediction_result.txt: Output file containing prediction results

Building the Project

Linux

# Compile PSNN
g++ -std=c++17 PSNN.cpp -o PSNN -I./onnxruntime-linux-x64-gpu-1.21.1/include -L./onnxruntime-linux-x64-gpu-1.21.1/lib -lonnxruntime

# Compile tester
g++ -std=c++17 tester.cpp -o tester

Windows

# Compile PSNN
cl /std:c++17 PSNN.cpp /Fe:PSNN.exe /I"path\to\onnxruntime\include" /link "path\to\onnxruntime\lib\onnxruntime.lib"

# Compile tester
cl /std:c++17 tester.cpp /Fe:tester.exe

Usage

Running a Prediction

1. Create input data in sharedData.txt or use the tester to generate sample data
2. Run the prediction:

# Linux
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:./onnxruntime-linux-x64-gpu-1.21.1/lib
./tester

# Windows
.\tester.exe

Input Data Format

The input data file (sharedData.txt) uses a comma-separated format:

FeatureName1,FeatureValue1
FeatureName2,FeatureValue2
...

Output Format

The prediction results are stored in prediction_result.txt:

Class 0,Class 1,Class 2,
0.628235,0.326715,0.0450506,
Predicted: Class 0 with 62.8235% confidence

Integration with Other Applications

The system is designed to be integrated with other applications through file-based communication:

1. External application generates input data in sharedData.txt
2. External application runs tester (or uses the functionality in tester.cpp)
3. External application reads prediction results from prediction_result.txt

Model Details

The prediction model (RDP_TripleNN.onnx) is a neural network that classifies inputs into three recombinant classes. The model expects standardized input features.

Data Processing Pipeline

1. Read raw input data from sharedData.txt
2. Drop features with no variance
3. Standardize features using pre-calculated means and standard deviations
4. Run inference through the ONNX model
5. Generate classification results and save to prediction_result.txt

License

MIT

Acknowledgements

• ONNX Runtime for providing the neural network execution environment# RdpLink