Raw code usage:

• Preprocessing (Seurat_code)

• Removing low-quality features and cells

  • Use "1_pre_analysis_Seurat3.R" (or your own QC pipeline)

• Analysis (Seurat_code)

• Normalize and visualize with Seurat for independet samples (or other analysis pipeline that generates at least normalized matrix as output). If more than one sample is present for a given omic first integrate them as usual pipelines requires.

  • Use "2_analysis_Seurat3.R" (or your own Normalization pipeline)

    Normalization method: Different normalization methods can enhance different LIBRA applications such as - by now - integration or prediction. As an example Seurat SCT method has shown similar integration performance but a significant decrease on prediction power.

• LIBRA (Libra_code)

• Use previous Analysis step generated .RData as input for LIBRA neural network

  • Use "LIBRA.R" for networks training
  • Use "Metrics_LIBRA.R" for additional quality metrix computation

    Outputs: Different outputs generated during the training will be stored in the working directory.

• Use supp_code.R for Python fine-tune full list of models PPJI computation employed on LIBRA fine-tune.