Spatially-mapped, genome-wide gene expression atlases of the brain are very valuable to study the genetic contribution to the anatomical organization of the brain. However, given the high-dimensionality of these atlases there is need for dimensionality reduction methods that can summarize both local and global relationships in the data to allow informative visualizations. We quantitatively assessed the performance of different dimensionality reduction techniques in separating neuroanatomical regions in low-dimensional (2D) embeddings of the mouse and human brains. We show that t-distributed Stochastic Neighbor Embedding (t-SNE) produces consistent embedding across 6 human brains from the Allen Human Brain Atlas as well as between the sagittal and coronal sections of the Allen Mouse Brain Atlas. We used low-dimensional embeddings to analyze the contribution of different cell-type markers in determining the structural organization of the mammalian brain.
- Ahmed Mahfouz
- Martijn van de Giessen
- Laurens van der Maaten (TU Delft)
- Sjoerd Huisman
- Marcel J.T. Reinders
- Mike Hawrylycz (Allen Institute for Brain Science)
- Boudewijn P.F. Lelieveldt
- Mahfouz A, van de Giessen M, van der Maaten L, Huisman SMH, Reinders MJT, Hawrylycz M, Lelieveldt BPF (2015) Visualizing the spatial gene expression organization in the brain through non-linear similarity embeddings. Methods 73:79–89.