Learning Patient Similarity from Genomics for Precision Oncology

This repository contains the scripts used in the model and analyses described in the manuscript titled: "Learning Patient Similarity from Genomics for Precision Oncology".

Shady M, Reardon B, Jiang S, Pimenta E, O'Meara T, Park J, Kehl KL, Elmarakeby HA, Sunyaev SR, Van Allen EM. Learning Patient Similarity from Genomics for Precision Oncology. medRxiv [Preprint]. 2025 Dec 18:2025.12.17.25342480. doi: 10.64898/2025.12.17.25342480. PMID: 41445600; PMCID: PMC12723981.

Code

Code used in data processing, model training, and evaluation is available in this repo, and should be run in Python (version >= 3.9.12)
Note that data required for portions of this analysis constitute protected health information for DFCI patients and therefore cannot be made publicly available. Researchers with DFCI appointments and Institutional Review Board (IRB) approval can access the data on request. For external researchers, access would require collaboration with the authors and eligibility for a DFCI appointment per DFCI policies.
Deidentified genomic data are available for DFCI patients through AACR's Project GENIE. Deidentified clinical data corresponding to PRISSMM annotations, pathologic information, imaging information, and medical oncologist evaluation for some DFCI patients is publicly available on cBioPortal and Synapse through the AACR Project GENIE Biopharmaceutical Consortium.
MSK-CHORD data is publicly available and was obtained through cBioPortal.

Python Dependencies

import argparse
import os
import yaml
import tqdm  
import pandas # version 1.4.2
import numpy # version 1.22.3
import scipy # version 1.10.0
import matplotlib # version 3.5.1
import seaborn # version 0.11.2
import comut # version 0.0.3 
import lifelines # version 0.27.7
import survive # version 0.3 
import pytorch # version 1.12.1
import scikit-learn # version 1.0.2
import tensorboard # version 2.9.0     

Operating Systems

The code is supported for macOS and Linux and has been tested on the following systems:

macOS Big Sur Version 11.6
macOS Sonoma Version 14.8.4
Linux Ubuntu 18.04 LTS

Hardware Requirements

Model training was performed on Google Cloud Platform (GCP) with the following configuration:

Compute instance: c2-standard-60 (60 vCPUs, 240 GB Memory)
CPU platform: Intel Cascade Lake
Disk: 200 GB SSD persistent disk
Operating System: Ubuntu 18.04 LTS (Linux)

License

GNU GENERAL PUBLIC LICENSE (Version 2)