Unsupervised Learning of Biological Networks using Graph Autoencoders
Track: Tiny Paper Track
Keywords: Graph Autoencoders, Unsupervised Learning, Biological Networks, Complex Systems, Network Biology
TL;DR: We propose the use of graph autoencoders for unsupervised learning of biological networks, achieving a compression ratio of 10:1 on a small dataset of 100 biological networks.
Abstract: We propose the use of graph autoencoders for unsupervised learning of biological networks. Graph autoencoders are a type of neural network that can learn to represent complex graph-structured data in a lower-dimensional space. We use a graph autoencoder to learn a representation of the biological network that captures the complex relationships between nodes and edges. Our preliminary results show that the graph autoencoder can learn a meaningful representation of the biological network that captures the complex relationships between nodes and edges. We achieve a compression ratio of 10:1 on a small dataset of 100 biological networks.
CompressionRatio=
CompressedDataSize
OriginalDataSize
Our results suggest that graph autoencoders can provide a meaningful representation of biological networks that captures the complex relationships between nodes and edges. We believe that further exploration of graph autoencoders can lead to new insights and understanding of biological systems.
GraphAutoencoder=Encoder×Decoder
We plan to further explore the potential of graph autoencoders for unsupervised learning of biological networks by investigating more complex graph autoencoder architectures, using larger datasets, and evaluating the performance of graph autoencoders on downstream tasks such as disease diagnosis and protein function prediction.
Submission Number: 21
Loading