00_demo_hadaca3_framework.Rmd

---
title: "Demo HADACA3 Framework"
author: "Hugo Barbot and Florent Chuffart "
date: "`r Sys.Date()`"
output: 
  rmarkdown::html_document:
    toc: true
    toc_float: true
    toc_depth: 3
    number_sections: true
---

```{r, include=(FALSE)}
knitr::opts_chunk$set(collapse=TRUE, comment="#>", fig.width=9, fig.height=6, eval=TRUE, echo=TRUE, results="verbatim")
options(knitr.duplicate.label = "allow")
```

# Load data

Demo adapted from baseline_teamF and teamF_Source_prior_known_feature (attachment). Scripts are located here : https://gricad-gitlab.univ-grenoble-alpes.fr/hadaca3/hadaca3_private/-/tree/main/baselines/attachement?ref_type=heads, and were copied in the `src` folder. 

Data were downloaded from the HADACA3 challenge starting kit. 

```{r}
ref = readRDS("00_demo_data/data/reference_pdac.rds")
mix = readRDS("00_demo_data/data/mixes1_SBN5_pdac.rds")
source("scr/teamF_Source_prior_known_features.R")
```

# Block Preprocessing (PP)

Takes as input a multimodal .txt.gz, outputs a multimodal txt.gz

```{r}
program_blockPP <- function(mix = NULL, ref = NULL, 
                    ...) {

   # Normalize input matrices

    ref$ref_bulkRNA = normalize_matrix(ref$ref_bulkRNA)
    ref$ref_met = normalize_matrix(ref$ref_met)


  
 multi_data = list(mix = mix,
             ref = ref)
  
  return(multi_data) 
}

PP_data = program_blockPP(mix = mix,
                          ref = ref)
```

# Block Feature selection (FS)

Takes as input a multimodal .txt.gz, outputs a multimodal txt.gz

```{r}

program_blockFS <- function(multi_data) {
   
   drop_null_ref_cols <- function(ref_matrix){
    non_null_rows = apply(ref_matrix != 0,MARGIN = 1, FUN = any, simplify = TRUE)
    return(ref_matrix[non_null_rows,])
   }

  ref_matrix = multi_data$ref$ref_bulkRNA
  multi_data$ref$ref_bulkRNA <- drop_null_ref_cols(ref_matrix)
  
  return(multi_data) 
}

FS_data = program_blockFS(multi_data = PP_data)
```

# Block SPLIT (SP)

```{r}

program_blockSP <- function(multi_data) {
   
   uni_data_RNA = list(mix = multi_data$mix$mix_rna,
                       ref = list(bulk = multi_data$ref$ref_bulkRNA,
                                  scRNA =  multi_data$ref$ref_scRNA))
   
      
   uni_data_met = list(mix = multi_data$mix$mix_met,
                       ref = list(bulk = multi_data$ref$ref_met))
   
  uni_data = list(RNA =  uni_data_RNA,
                  met = uni_data_met)
  return(uni_data) 
}

SP_data = program_blockSP(multi_data = FS_data)
```


# Block Deconvolution (DE)

Takes as input a multimodal mix.txt.gz and a ref.txt.gz, output a prediction.txt.gz


```{r}
uni_data = SP_data$RNA

program_blockDE <- function(uni_data) {

    ## RNA:
  #if ( !( is.null(x = mix_rna) ) ) {
    
    idx_feat = intersect(rownames(uni_data$mix), rownames(uni_data$ref$bulk))
    uni_data$mix = uni_data$mix[idx_feat,]
    uni_data$ref$bulk = uni_data$ref$bulk[idx_feat,]
     uni_pred = get_prop_nnls(uni_data$mix,  uni_data$ref$bulk)
     

  

  
  return(uni_pred) # MAG : je ne sais pas quoi sortir 
}


pred_RNA = program_blockDE(uni_data = SP_data$RNA)
pred_met = program_blockDE(uni_data = SP_data$met)
```

# Block Late integration (LI)



Takes as input a a multimodal prediction.txt.gz, outputs an integrated prediction


```{r}
program_blockLI <- function(pred1, pred2) { 

        multi_pred <- (pred1 + pred2) / 2

  
  return(multi_pred) # MAG : je ne sais pas quoi sortir 
}

final_pred = program_blockLI(pred_RNA, pred_met)
```

# Session Information

```{r, results="verbatim"}
sessionInfo()
```