# Statistical Modelling of AP-MS Data (SMAD)
This R package implements statistical modelling of affinity purification–mass spectrometry (AP-MS) data to compute confidence scores to identify *bona fide* protein-protein interactions (PPI).
## Installation
The development version can be installed through github:
```{r}
devtools::install_github(repo="zqzneptune/SMAD")
library(SMAD)
```
## Input Data
A demo data.frame was provided as a hint how the input data should strcutured in order to run the scoring functions:
```{r}
data(TestDatInput)
colnames(TestDataInput)
[1] "idRun" "idBait" "idPrey" "countPrey" "lenPrey"
```
|idRun|idBait|idPrey|countPrey|lenPrey|
|-----|:----:|:----:|:-------:|-------|
|Unique ID of one AP-MS run|Bait ID|Prey ID|Prey peptide count|Protein sequence length of the prey|
In case of duplcates, a suffix or prefix of e.g. "A", "B" could be added to **idRun** in order to make **"idRun-idBait"** combination unique to each replicate.
## Run scoring
### 1. CompPASS
Comparative Proteomic Analysis Software Suite (CompPASS) is based on spoke model. This algorithm was developed by Dr. Mathew Sowa for defining the human deubiquitinating enzyme interaction landscape [(Sowa, Mathew E., et al., 2009)][1]. The implementation of this algorithm was inspired by Dr. Sowa's [online tutorial][2]. The output includes Z-score, S-score, D-score and WD-score. In its implementation in BioPlex 1.0 [(Huttlin, Edward L., et al., 2015)][3] and
BioPlex 2.0 [(Huttlin, Edward L., et al., 2017)][4], a naive
Bayes classifier that learns to distinguish true interacting proteins from
non-specific background and false positive identifications was included in the
compPASS pipline. This function was optimized from the [source code][5].
The input data.frame, *datInput*, should include:**idRun**, **idBait**, **idPrey** and **countPrey**.
```{r}
datScore <- CompPASS(datInput)
```
### 2. DICE
The Dice coefficient is used to score the interaction scores across prey pair-wise combinations, which was proposed by [(Bing Zhang et al., 2008)][9]
The input data.frame, *datInput*, should include:**idRun** and **idPrey**.
```{r}
datScore <- DICE(datInput)
```
### 3. Hart
Hart scoring algorithm is based on a hypergeometric distribution error model [(Hart et al., 2007)][6].
The input data.frame, *datInput*, should include:**idRun** and **idPrey**.
```{r}
datScore <- Hart(datInput)
```
### 4. HGScore
HGScore algorithm is based on a hypergeometric distribution error model [(Hart et al., 2007)][6] with incorporation of NSAF [(Zybailov, Boris, et al., 2006)][7]. This algorithm was first introduced to predict the protein complex network of Drosophila melanogaster [(Guruharsha, K. G., et al., 2011)][8]. This scoring algorithm was based on matrix model.
The input data.frame, *datInput*, should include:**idRun**, **idPrey**, **countPrey** and **lenPrey**.
```{r}
datScore <- HG(datInput)
```
### 5. PE
PE incorporated both spoke and matrix model as repored in [(Sean R. Collins, et al., 2007)][10].
The input data.frame, *datInput*, should include:**idRun**, **idBait** and **idPrey**.
```{r}
datScore <- PE(datInput)
```
## License
MIT @ Qingzhou Zhang
[1]: https://doi.org/10.1016/j.cell.2009.04.042
[2]: http://besra.hms.harvard.edu/ipmsmsdbs/cgi-bin/tutorial.cgi
[3]: https://doi.org/10.1016/j.cell.2015.06.043
[4]: https://www.nature.com/articles/nature22366
[5]: https://github.com/dnusinow/cRomppass
[6]: https://doi.org/10.1186/1471-2105-8-236
[7]: https://doi.org/10.1021/pr060161n
[8]: https://doi.org/10.1016/j.cell.2011.08.047
[9]: https://doi.org/10.1093/bioinformatics/btn036
[10]: https://doi.org/10.1074/mcp.M600381-MCP200
