# Master’s Thesis Proposal
Edit Proposal on Overleaf: https://www.overleaf.com/2497778233nqxqdzkbmbqb
Github repository: https://github.com/walzph/gpu-accelerated-medecom

### What is DNA methylation?
- robust mehtods exist for detection and quantification of methylation data
- mirrors functional state of a cell
- characteristic methylation profile (methylome)
- occur at CpG dinucleotides
- cell-type specific patterns
- influenced by
- individual genetic constitution, gender, age, environment
- important for comparative studies

### Why MeDeCom ?
- can decompose raw methylome data into latend **DNA methylation components (LMCs)**
- computational framework using:
- regularized non-negative matrix factorization
- biologically motivated regularizer
- favors LMC with per-CpG values close to one or zero (i.e. methylated or unmethylated)
- key element for accurate estimation
- still works if only mixtures of different cell types / no purified references available
- Matrix D represents CpG sites

#### Resources
> Website: http://public.genetik.uni-sb.de/medecom/
> Paper: https://genomebiology.biomedcentral.com/articles/10.1186/s13059-017-1182-6
> Current implementation based on R: https://github.com/lutsik/MeDeCom
> **Are some parts already implemented in C++ ?**

#### Internal Links:
[[Matrix Facorization]]
[[Singular Value Decomposition]]
[[s13059-017-1182-6.pdf]]

## Optimization Approaches

### Matrix Factorization (fine-grained)
GPU acceleration of MF: https://dorukkilitcioglu.github.io/2018/12/21/cu2rec.html
Cu2Rec: https://github.com/nickgreenquist/cu2rec

### Run orchestration (coarse-grained)