We develop a systematic combinatorial design strategy that uses machine learning to prioritise the most promising targeted drug combinations for relapsed/refractory AML (RR-AML) patients using single-cell transcriptomics and single-agent response profiles measured in primary patient samples. By utilizng the established target-based Normalized Single-cell Enrichment (t-NSE) score, we can quantitatively compare the co-inhibition effects of drug combinations among various cell types and prioritize combinations that exhibit high synergy and potency in co-inhibiting AML cells, while showing non-synergistic effects in non-malignant cells. The following figure illustrates the workflow of the drug combination prediction and testing pipeline (created using Biorender).
To install all dependencies, the version of R should be >= 4.0.5. The required packages and their recommended versions are as follows:
[1] ModelMetrics_1.2.2.2 caret_6.0-92 parallel_4.0.5 openxlsx_4.2.6.1
[4] xgboost_1.5.1.1 Seurat_4.1.1 ggplot2_3.3.6 Biobase_2.64.0
[7] readr_2.1.2 copykat_1.0.8 HGNChelper_0.8.1
[10] dplyr_1.0.9 lhs_1.2.0 ParamHelpers_1.14.1
For any questions please contact Yingjia Chen ([email protected])
Code copyright Prediction of selective and synergistic drug combinations for relapsed AML
License https://github.com/yingjchen/RR-AML/blob/main/LICENSE
The original version of the code for developing XGBoost models was forked from https://github.com/IanevskiAleksandr/scComb. Many thanks to Aleksandr Ianevski for making his code available.