---
title: "CytoMethIC Quick Start"
shorttitle: "CytoMethIC Quick Start"
date: "`r BiocStyle::doc_date()`"
package: CytoMethIC
output: BiocStyle::html_document
fig_width: 6
fig_height: 5
vignette: >
    %\VignetteEngine{knitr::knitr}
    %\VignetteIndexEntry{CytoMethIC User Guide}
    %\VignetteEncoding{UTF-8}
    
---
# Installation

To use CytoMethIC, you need to install the package from Bioconductor. If you
don't have the BiocManager package installed, install it first:

```{r, message=FALSE, warning=FALSE, cache=TRUE, eval=FALSE}
if (!requireNamespace("BiocManager", quietly = TRUE)) {
  install.packages("BiocManager")
}
if (!requireNamespace("CytoMethIC", quietly = TRUE)) {
  BiocManager::install("CytoMethIC")
}
```
# Introduction

`CytoMethIC` is a comprehensive package that provides model data and functions
for easily using machine learning models that use data from the DNA methylome
to classify cancer type and phenotype from a sample. The primary motivation for
the development of this package is to abstract away the granular and
accessibility-limiting code required to utilize machine learning models in
R. Our package provides this abstraction for RandomForest, e1071 Support
Vector, Extreme Gradient Boosting, and Tensorflow models. This is paired with
an ExperimentHub component, which contains our lab's models developed for
epigenetic cancer classification and predicting phenotypes. This includes CNS tumor
classification, Pan-cancer classification, race prediction, cell of origin
classification, and subtype classification models.

```{r, cyto2, message=FALSE, warning=FALSE}
library(CytoMethIC)
library(ExperimentHub)
library(sesame)
sesameDataCache()
```

# Data from ExperimentHub

For these examples, we'll be using models from ExperimentHub and a sample from
sesameData.

```{r cyto3, result="asis", echo=FALSE}
library(knitr)
kable(cmi_models[, c("ModelID", "PredictionLabelDescription")],
  caption = "CytoMethIC supported models"
)
```

# Pan-Cancer type classification

The below snippet shows a demonstration of the model abstraction working on
random forest and support vector models from CytoMethIC models on
ExperimentHub.

```{r cyto4, message=FALSE}
cmi_predict(sesameDataGet("HM450.1.TCGA.PAAD")$betas, ExperimentHub()[["EH8395"]],
    lift_over=TRUE)
cmi_predict(sesameDataGet("HM450.1.TCGA.PAAD")$betas, ExperimentHub()[["EH8396"]],
    lift_over=TRUE)
```

# Pan-Cancer subtype classification

The below snippet shows a demonstration of the cmi_predict function working to
predict the subtype of the cancer.

```{r cyto5, message=FALSE}
cmi_predict(sesameDataGet("HM450.1.TCGA.PAAD")$betas, ExperimentHub()[["EH8422"]])
```

# Ethnicity classification

The below snippet shows a demonstration of the cmi_predict function working to
predict the ethnicity of the patient.

```{r cyto6, message=FALSE}
cmi_predict(sesameDataGet("HM450.1.TCGA.PAAD")$betas, ExperimentHub()[["EH8421"]])
```
# Pan-Cancer COO classification

The below snippet shows a demonstration of the cmi_predict function working to
predict the cell of origin of the cancer.

```{r cyto7, message=FALSE}
cmi_predict(sesameDataGet("HM450.1.TCGA.PAAD")$betas, ExperimentHub()[["EH8423"]])
```

# Ethnicity From GitHub Link

In addition to ExperimentHub Models, this package also supports using models
from GitHub URLs. Note that https://github.com/zhou-lab/CytoMethIC_models will
be the most frequently updated public repository of our lab's classifiers.

```{r cyto8}
base_url <- "https://github.com/zhou-lab/CytoMethIC_models/raw/main/models"
cmi_model <- readRDS(url(sprintf("%s/Race3_rfcTCGA_InfHum3.rds", base_url)))
betas <- openSesame(sesameDataGet("EPICv2.8.SigDF")[[1]])
cmi_predict(betas, cmi_model, lift_over=TRUE)
```

```{r}
sessionInfo()
```