# Dimension Reduction
> [!quote] Linear vs Non-Linear Reductions
> Intuitively, you can think about linear transformations as shifting and stretching the data, meanwhile non-linear transformations will make more dramatic changes on data, such as making it “inside out”.
https://en.wikipedia.org/wiki/Curse_of_dimensionality
## Types of Dimension Reduction Techniques
### Supervised vs Unsupervised
> See also:
> - [[Machine Learning]]
**Supervised:**
- LDA
**Unsupervised:**
- PCA
### Linear vs Non-Linear
#### Linear Transformations
> See also:
> - [[Linear Algebra]]
![[Pasted image 20240124164440.png|275]]
[[Principal Component Analysis (PCA)]]
- Can reduce data to a lower dimension while keeping the information (variance) in our data.
- Converts data into an eigenvector and eigenvalue
- Tries to reduce dimensionality by maximizing variance in the data
#### Non-Linear Transformations
![[Pasted image 20240124164628.png|327]]
**Kernal PCA (kPCA)**
- An extension of PCA by introducing a
**t-distributed Stochastic Neighbor Embedding (t-SNE)**
- Tries to do the same as PCA by keeping similar data points together (and dissimilar data points apart) in both higher and lower dimensions
- [StatQuest: t-SNE, Clearly Explained](https://www.youtube.com/watch?v=NEaUSP4YerM)
**Uniform Manifold Approximation and Projection (UMAP)**
- Supposedly very similar to t-SNE (considered a subset by some)
- [ ] autoencoders
## Applications of Dimension Reduction
### Differential Gene Expression
A gene is declared as differentially expressed (DE) if an observed difference or change in read counts or expression levels between two experimental conditions is statistically significant.
- [NIH Paper on an Approach to DE](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4827276/)
Cells can be clustered based on their PCA scores, with each PC essentially representing a ‘metafeature’ that combines information across a correlated feature set.
## Visualizing Dimension Reduction
---
https://www.reddit.com/r/datascience/comments/wy1rmk/pca_vs_umap_vs_tsne_on_a_very_layman_level_what/
PCA is very useful