To study the drug resistance problem caused by transporters, we leveraged multiple large-scale public data sets of drug sensitivity, cell line genetic and transcriptional profiles, and gene silencing experiments. Through systematic integration of these data sets, we built various machine learning models to predict the difference between cell viability upon drug treatment and the silencing of its target across the same cell lines. More than 50% of the models built with the same data set or with independent data sets successfully predicted the testing set with significant correlation to the ground truth data. Features selected by our models were also significantly enriched in known drug transporters annotated in DrugBank for more than 60% of the models. Novel drug-transporter interactions were discovered, such as lapatinib and gefitinib with ABCA1, olaparib and NVPADW742 with ABCC3, and gefitinib and AZ628 with SLC4A4.
Fig: Boxplot of pairwise drug similarity scores targeting the same target in each data set. The similarity score is represented by the negative log10 p-value of the pairwise correlations computed based on their sensitivities across cancer cell lines.
Furthermore, we identified ABCC3, SLC12A7, SLCO4A1, SERPINA1, and SLC22A3 as potential transporters for erlotinib, three of which are also significantly more highly expressed in patients who were resistant to therapy in a clinical trial.
Shen, Y., Yan, Z. Systematic prediction of drug resistance caused by transporter genes in cancer cells. Sci Rep 11, 7400 (2021). https://doi.org/10.1038/s41598-021-86921-9