MLquiades

2026/07/13

(pronounced em-el-kee-ah-days)

This package takes in bulk RNA cancer cell line sequencing data (processed from raw fastqs to TPM counts using STAR) and GDSC1 IC50 drug sensitivity scores for palbociclib to build and evaluate four machine learning models.

These models include: neural net with hyperband parameter tuning, random forest (ensemble), ridge classifier (L2 regularization), and SVM (L1 regularization).

There are avenues for feature, in this case gene, selection. They include: only CDK4 and CDK6 related genes (the target for palbociclib); only CDK4, CDK6 and cancer genes with well-known somatic mutations (COSMIC); and a Pearson correlation method that keeps only the genes that have rho>=.3 value with the IC50 score in the training dataset only.

Installation

Ubuntu and MacOS are supported. Windows is not currently supported.

git init
git clone git@github.com:HorvathLab/mlquiades.git
cd mlquiades

If you use a different package manager, install packages to your environment from pyproject.toml

Usage (default)

If using uv

uv venv
# activate the venv: your command may differ for a different shell
source .venv/bin/activate
uv python pin 3.10
zenodo_get 21893525 -o sample_data #https://zenodo.org/records/21893525
tar -xvf sample_data/palbociclib_gex_isos.gz -C sample_data
uv run src/mlquiades/main.py --a sample_data --b output_dir

If not using uv

pip install -e .
zenodo_get 21043540 -o sample_data #https://zenodo.org/records/21842954
tar -xvf sample_data/palbociclib_gex_isos.gz -C sample_data
python src/mlquiades/main.py --a sample_data --b output_dir

Output

└── output
    ├── cdk4_6_cancer_both
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_cancer.png
    │   └── plt_rocauc_all_cdk4_6_cancer.png
    ├── cdk4_6_cancer_gex
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_cancer.png
    │   └── plt_rocauc_all_cdk4_6_cancer.png
    ├── cdk4_6_cancer_isoforms
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_cancer.png
    │   └── plt_rocauc_all_cdk4_6_cancer.png
    ├── cdk4_6_genes_both
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_genes.png
    │   └── plt_rocauc_all_cdk4_6_genes.png
    ├── cdk4_6_genes_gex
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_genes.png
    │   └── plt_rocauc_all_cdk4_6_genes.png
    ├── cdk4_6_genes_isoforms
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_cdk4_6_genes.png
    │   └── plt_rocauc_all_cdk4_6_genes.png
    ├── data_split.csv
    ├── data_split.png
    ├── pearson_both
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_pearson.png
    │   └── plt_rocauc_all_pearson.png
    ├── pearson_gex
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_pearson.png
    │   └── plt_rocauc_all_pearson.png
    ├── pearson_isoforms
    │   ├── evaluation_df_acc.csv
    │   ├── evaluation_df.csv
    │   ├── evaluation_df_rocauc.csv
    │   ├── plt_accuracy_all_pearson.png
    │   └── plt_rocauc_all_pearson.png
    ├── report_both.html
    ├── report_both.md
    ├── report_gex.html
    ├── report_gex.md
    ├── report_isoforms.html
    └── report_isoforms.md

Authors

Horvath Lab + Joe Goldfrank

George Washington University

McCormick Genomic and Proteomic Center (MGPC) & Department of Computer Science

Questions? siera.martinez@gwu.edu