- Some new work from myself, @alexfornito.bsky.social, and @garedaba.bsky.social where we investigate the performance of generative Network Models (GNMs) 🧠🌐🤖. A 🧵 (1/6) www.biorxiv.org/content/10.1...
- GNMs create synthetic networks based on mechanisms/constraints that are believed to shape brain networks. The resulting networks replicate topological properties of real brain networks, but not topographical properties (i.e., spatial embedding of topology) (2/6)
- We find this is because GNMs cannot accurately capture long-range connections, these connections are needed to define the precise topography we observe in the empirical data (3/6)
- In addition, we show how a common way of evaluating GNMs is very sensitive to differences in short-range, but not long-range, connections. This has implications for how to select best-fitting models (4/6)
- So where does this work leave GNMs? In fairness, they were originally designed to capture topology, not topography. We wanted to push the models to their limit to identify where they can be improved (5/6)Nov 19, 2024 16:07
- As the brain's spatial embedding is a key aspect of its organisation and function, we believe GNMs should aim to capture this going forward. Features like heterochronicity may improve GNMs ability in this regard. Check out more of our ideas in the paper! 😊 (6/6) www.biorxiv.org/content/10.1...
