Uniform Manifold Approximation and Projection (UMAP)-based sorting enhances detection of low-firing rate neurons and boosts mutual information.
<p><b>(A)</b> Cumulative average firing rate distributions for UMAP-based and principal component analysis (PCA)-based spike sorting reveal that UMAP identifies more low-firing neurons, reflecting greater sensitivity to sparse firing. <b>(B)</b> Anatomical maps illustra...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | , , , , , , , , , , |
| 出版: |
2025
|
| 主题: | |
| 标签: |
添加标签
没有标签, 成为第一个标记此记录!
|
| 总结: | <p><b>(A)</b> Cumulative average firing rate distributions for UMAP-based and principal component analysis (PCA)-based spike sorting reveal that UMAP identifies more low-firing neurons, reflecting greater sensitivity to sparse firing. <b>(B)</b> Anatomical maps illustrate three cortical regions: ventral premotor cortex (VPC, green), secondary somatosensory cortex (S2, red), and dorsal prefrontal cortex (DPC, blue). Dotted lines in the lightly shaded oval above S2 indicate recording sites in deeper cortical layers—areas involved in higher-level processing and decision-making. <b>(C–E)</b> Comparisons of PCA and UMAP sorting across these regions. Left: UMAP consistently recovers a larger number of low-firing neurons based on cumulative firing rate distributions: 134 (UMAP) and 60 (PCA) neurons were identified in S2; 228 (UMAP) and 145 (PCA) in VPC; 129 (UMAP) and 96 (PCA) in DPC. Center: It also yields higher overall neuron counts from the same sessions: average number of neuron pairs in S2 from about 4 to almost 10; VPC from about 5 to almost 9; DPC from about 6 to almost 8. Right: UMAP reveals greater mutual information (MI) compared to PCA, indicating stronger information decoding across all brain regions. In summary, UMAP effectively increases neuron recovery, captures low-firing activity, and enhances the information gained from cognitive task data. The neuronal activity recorded during the foreperiod of the cognitive tasks, used for the cumulative distributions, is available at [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003527#pbio.3003527.ref044" target="_blank">44</a>]; the single neurons from different sessions used for the mutual information calculations can be found at [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003527#pbio.3003527.ref043" target="_blank">43</a>]; and the corresponding analysis code is available at [<a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.3003527#pbio.3003527.ref049" target="_blank">49</a>].</p> |
|---|