Subtype Classification of Tumors and Derived Lab Grown Models

Molecular subtyping using the TMP Toolkit

Table of contents

• Quickstart Guide
• Download Data from Manifest File Using the GDC Client
• Run Processing Pipeline
• Sample Subtype Classification using Gene Expression Data
• Sample Subtype Classification using DNA Methylation Data

Quickstart Guide

Setup

Install requirements - detailed instructions are found on the Requirements page:

1. Install Python 3+
2. Install GDC Data Transfer Tool Client

Ensure that steps are completed on the Requirements page - (includes creating working environment, signining in, and manually downloading required data)

Download Data from Manifest File Using the GDC Client

Download Gene Expression Data

bash scripts/gdc_download.sh PAAD

This will create subfolders in data-raw/GEXP` and place GDC molecular matrices here.

Options for cancer cohort includes ALL, BLCA, BRCA, COADREAD, ESO, HNSC, KID, LGGGBM, LIHCCHOL, LUNG, OV, PAAD, SARC, SKCM, UCEC

For more details on each cancer cohort option see Cohort Options Page

Run Processing Pipeline

Example shown for running PAAD cohort

bash scripts/process.sh PAAD data/prep

Creates file data/prep/<CANCER>_GEXP/<CANCER>_GEXP_prep2_<TYPE>.tsv that is prepped for distance calculations

Options for cancer cohort includes ALL, BLCA, BRCA, COADREAD, ESO, HNSC, KID, LGGGBM, LIHCCHOL, LUNG, OV, PAAD, SARC, SKCM, UCEC

For more details on each cancer cohort option see Cohort Options Page

Sample Subtype Classification Using Gene Expression Data

The goal of this analysis is to get cancer subtype predictions for HCMI samples (organoids, cell cultures , xenografts, etc). To accomplish this we will use the top performing pre-trained machine learning models (dockerized TMP models that were trained using TCGA data that has been pre-proccessed). Specifically we are interested in using gene expression from the HCMI samples and eventually compare primary tumors to their corresponding models (organoids, cell cultures , xenografts, etc).

The TMP models (pre-trained models) are specific to TCGA cancer cohorts (TCGA abbreviations), therefore we will split HCMI data into TCGA cancer cohorts(based on sample metadata).

Run gene expression classifier pipeline:

# where specify cancer, tumor-file, model-file, transformed-dir
bash scripts/run_classify_GEXP.sh \
    PAAD \
    data/prep/PAAD_GEXP/PAAD_GEXP_prep2_Tumor.tsv \
    data/prep/PAAD_GEXP/PAAD_GEXP_prep2_Model.tsv \
    data/classifier_gexp/ml_ready_qrank

Results can found in data/classifier_gexp/ml_predictions_qrank/combo/HCMI_TMPsubtype_qRank_<CANCER>.tsv

Note: LUNG (includes LUAD and LUSC), ESO (includes GEA and ESCC) during transformation and classification, then is merged in post-classification summary

Sample Subtype Classification Using DNA Methylation Data

The goal of this analysis is to get cancer subtype predictions for HCMI samples (organoids, cell cultures , xenografts, etc). To accomplish this we will use the top performing pre-trained machine learning models (dockerized TMP models that were trained using TCGA data that has been pre-proccessed). Specifically we are interested in using gene expression from the HCMI samples and eventually compare primary tumors to their corresponding models (organoids, cell cultures , xenografts, etc).

The TMP models (pre-trained models) are specific to TCGA cancer cohorts (TCGA abbreviations), therefore we will split HCMI data into TCGA cancer cohorts(based on sample metadata).

Run DNA methylation classifier pipeline:

# where specify cancer, tumor-file, model-file, transformed-dir
bash scripts/run_classify_METHYL.sh \
    SKCM \
    data/classifier_methyl/processed/20231211_HCMI_TMP_subtype_prediction_feature_matrix_SKCM.tsv

Results can found in data/classifier_methyl/ml_predictions/combo/HCMI_METH_TMPsubtypes.<CANCER>.tsv

Note: LUNG (includes LUAD and LUSC), ESO (includes GEA and ESCC) during transformation and classification, then is merged in post-classification summary

Second Example for Combination Cohort

bash scripts/run_classify_METHYL.sh \
    LUNG \
    data/classifier_methyl/processed/20231211_HCMI_TMP_subtype_prediction_feature_matrix_LUNG.tsv