Biomarker discovery and patient stratification in pancreatic cancer using incomplete multi-omics data

This repository contains the data and scripts used in the paper "Biomarker discovery and patient stratification in pancreatic cancer using incomplete multi-omics data".

Prerequisites | Data download | Benchmark steps | Cluster analysis | PDAC classifications comparison | Omics Analysis | Predict

Prerequisites

To reproduce the results from our study, you will need the following software installed on your system: Python (3.11), Octave and R (version 4.0 or higher).

Clone this repository running:

git clone https://github.com/ocbe-uio/imoc_pdac.git
cd imoc_pdac

All required Python packages are listed in requirements.txt. Install them using:

pip install -r requirements.txt

Data download

To download the data used in this study, run the R script /data/download_data.R:

Rscript data/download_data.R

To process the data, run the script /data/data_preprocessing.py twice, once with complete_sample_set = False and complete_sample_set = True to generate the entire data used in the study.

python data/data_preprocessing.py

Benchmark steps

To obtain the results found in the /results/cluster_analysis/benchmarking_files folder, run the file /src/scripts/generating_indxs.py first, followed by /src/scripts/incomplete_algorithms_evaluation.py. The settings must be changed for every benchmark in the settings.py file (see below). Warning: it can take several hours to run every script.

python src/scripts/generating_indxs.py
python src/scripts/incomplete_algorithms_evaluation.py

First benchmark

best_combination = False
run_amputation = False
select_datasets = ['patients_with_all_views']
n_clusters = [2, 3, 4, 5]   # can be done for each number of clusters individually, e.g.: 2 clusters, 3 clusters, etc.
runs_per_alg = np.arange(10)
sampling = True

Second benchmark

best_combination = ["CNA", "Methylation"]
run_amputation = True
select_datasets = ['patients_with_all_views']
n_clusters = [2]
runs_per_alg = np.arange(10)
sampling = False

Final clusters

best_combination = ["CNA", "Methylation"]
run_amputation = False
select_datasets = ['all_patients']
n_clusters = [2]
runs_per_alg = np.arange(1024)
sampling = True

Cluster analysis

The analysis of the final clusters can be reproduced by running the jupyter notebooks in the /notebooks folder. This will generate the data used for the figures and supplementary data. The files

PDAC classifications comparison

To reproduce the analysis, results, and figures of the comparison of clusters with other PDAC classification taxonomies, open the .Rproj file in R. Then run the R script /notebooks/paper_clusters_analysis.R.

Omics analysis

To reproduce the omics analysis, open the .Rproj file in R. Then, run the R scripts in the /omics_analysis folder. This will generate the data used for the figures and supplementary data.

To reproduce the manuscript's figures and supplementary data, run the script /omics_analysis/05_Results_Figure.R

Use our model to stratify patients

We have released the CNA-based PDAC patient stratification model to facilitate the stratification of new samples and clinical translation. This model requires only the four selected copy number loci ('21q11.2', '17p12', '18q21.2', '9p21.3'), enabling direct application without the need for preprocessing or batch correction.

If you want to use our model to stratify patients, start loading the model in Python:

import pickle
with open("cna_rfmodel.pkl", "rb") as f:
    model = pickle.load(f)

Then, load your dataset. The possible values are:

• -2: Homozygous deletion (both chromosomes)
• -1: Heterozygous deletion (one chromosome)
• 0: Wild-type (no alteration)
• 1: Single copy gain (one chromosome)
• 2: Amplification (both chromosomes)

For this example, we will just create a random dataset with 5 patients:

import numpy as np
import pandas as pd
n_patients = 5
X = np.random.default_rng(42).integers(-2, 3, size=(n_patients, 4))
X = pd.DataFrame(X, columns=['21q11.2', '17p12', '18q21.2', '9p21.3'])

Finally, predict the cluster of your patients:

model.predict(X)