Frequently Asked Questions (FAQ)

Q1: What is BioNeuralNet?

A1: BioNeuralNet is a Python framework for integrating multi-omics data with Graph Neural Networks (GNNs). It provides tools for graph construction, clustering, network embedding, subject representation, and disease prediction.

Q2: What are the key features of BioNeuralNet?

A2: BioNeuralNet includes the following components:

  • Graph Construction: Build multi-omics networks using SmCCNet or custom adjacency matrices.

  • Graph Clustering: Identify meaningful communities with Louvain, Hybrid Louvain, or PageRank-based methods.

  • GNN Embedding: Learn node embeddings from biological graphs with GNNEmbedding.

  • Subject Representation: Integrate embeddings into omics data via GraphEmbedding for enhanced feature learning.

  • Disease Prediction: Use DPMON, an end-to-end pipeline that trains a GNN-based classifier.

Q3: How do I install BioNeuralNet?

A3: Install BioNeuralNet with:

pip install bioneuralnet

For GPU support, install PyTorch with CUDA from [PyTorch.org](https://pytorch.org/get-started/locally/). See Installation for full setup details.

Q4: Does BioNeuralNet support GPU acceleration?

A4: Yes. If you have a CUDA-compatible GPU, BioNeuralNet will automatically use it if torch.cuda.is_available() is True.

Q5: Can I use my own adjacency matrix instead of SmCCNet?

A5: Yes! If you have a precomputed adjacency matrix, you can pass it directly to GNNEmbedding or DPMON. SmCCNet is an optional tool for generating adjacency matrices.

Q6: How is DPMON different from other GNN models?

A6: DPMON is designed for multi-omics disease prediction. Unlike standard GNNs, it:

  • Jointly learns node embeddings and a classifier.

  • Leverages both local and global graph structures.

  • Integrates phenotype and clinical data alongside omics features.

Q7: Can I run GNNEmbedding without labeled data (unsupervised learning)?

A7: Yes! If you don’t provide labels, GNNEmbedding will still generate embeddings based on graph structure. For self-supervised learning (e.g., contrastive learning), you may need additional adaptation.

Q8: What clustering methods does BioNeuralNet support?

A8: BioNeuralNet provides:

  • Correlated Louvain: Clusters nodes based on omics similarity and phenotype correlation.

  • Hybrid Louvain: Iteratively refines clusters using PageRank expansion.

  • Correlated PageRank: Detects communities based on personalized PageRank scores.

Q9: Can I contribute new features or models?

A9: Yes! We welcome contributions. Fork the repository, add your module, and submit a pull request. Check our contribution guide.

Q10: What license is BioNeuralNet under?

A10: BioNeuralNet is released under the MIT License.

Q11: How do I report issues or request features?

A11: Open an issue on our GitHub repository: UCD-BDLab/BioNeuralNet.

Q12: Where can I find tutorials or example scripts?

A12: See Tutorials for step-by-step guides on graph construction, embeddings, subject representation, and disease prediction.