No abstract

Despite incredible progress in cancer treatment, therapy resistance remains the leading limiting factor for long−term survival. During drug treatment, several genes are transcriptionally upregulated to mediate drug tolerance. Using highly variable genes and pharmacogenomic data for acute myeloid leukemia (AML), we developed a drug sensitivity prediction model for the receptor tyrosine kinase inhibitor sorafenib and achieved more than 80% prediction accuracy. Furthermore, by using Shapley additive explanations for determining leading features, we identified AXL as an important feature for drug resistance. Drug−resistant patient samples displayed enrichment of protein kinase C (PKC) signaling, which was also identified in sorafenib−treated FLT3−ITD−dependent AML cell lines by a peptide−based kinase profiling assay. Finally, we show that pharmacological inhibition of tyrosine kinase activity enhances AXL expression, phosphorylation of the PKC−substrate cyclic AMP response element binding (CREB) protein, and displays synergy with AXL and PKC inhibitors. Collectively, our data suggest an involvement of AXL in tyrosine kinase inhibitor resistance and link PKC activation as a possible signaling mediator.

Therapy resistance remains one of the major challenges for cancer treatment that largely limits treatment benefits and patient survival. The underlying mechanisms that lead to therapy resistance are highly complicated because of the specificity to the cancer subtype and therapy. The expression of the anti-apoptotic protein BCL2 has been shown to be deregulated in T-cell acute lymphoblastic leukemia (T-ALL), where different T-ALL cells display a differential response to the BCL2-specific inhibitor venetoclax. In this study, we observed that the expression of anti-apoptotic BCL2 family genes, such as BCL2, BCL2L1, and MCL1, is highly varied in T-ALL patients, and inhibitors targeting proteins coded by these genes display differential responses in T-ALL cell lines. Three T-ALL cell lines (ALL-SIL, MOLT-16, and LOUCY) were highly sensitive to BCL2 inhibition within a panel of cell lines tested. These cell lines displayed differential BCL2 and BCL2L1 expression. Prolonged exposure to venetoclax led to the development of resistance to it in all three sensitive cell lines. To understand how cells developed venetoclax resistance, we monitored the expression of BCL2, BCL2L1, and MCL1 over the treatment period and compared gene expression between resistant cells and parental sensitive cells. We observed a different trend of regulation in terms of BCL2 family gene expression and global gene expression profile including genes reported to be expressed in cancer stem cells. Gene set enrichment analysis (GSEA) showed enrichment of cytokine signaling in all three cell lines which was supported by the phospho-kinase array where STAT5 phosphorylation was found to be elevated in resistant cells. Collectively, our data suggest that venetoclax resistance can be mediated through the enrichment of distinct gene signatures and cytokine signaling pathways.

Therapeutic resistance continues to impede overall survival rates for those affected by cancer. Although driver genes are associated with diverse cancer types, a scarcity of instrumental methods for predicting therapy response or resistance persists. Therefore, the impetus for designing predictive tools for therapeutic response is crucial and tools based on machine learning open new opportunities. Here, we present an easily accessible platform dedicated to Clear, Legible, Explainable, Transparent, and Elucidative (CLETE) yet wholly modifiable binary classification models. Our platform encompasses both unsupervised and supervised feature selection options, hyperparameter search methodologies, under-sampling and over-sampling methods, and normalization methods, along with fifteen machine learning algorithms. The platform furnishes a k-fold receiver operating curve (ROC) - area under the curve (AUC) and accuracy plots, permutation feature importance, SHapley Additive exPlanations (SHAP) plots, and Local Interpretable Model-agnostic Explanations (LIME) plots to interpret the model and individual predictions. We have deployed a unique custom metric for hyperparameter search, which considers both training and validation scores, thus ensuring a check on under or over-fitting. Moreover, we introduce an innovative scoring method, NegLog2RMSL, which incorporates both training and test scores for model evaluation that facilitates the evaluation of models via multiple parameters. In a bid to simplify the user interface, we provide a graphical interface that sidesteps programming expertise and is compatible with both Windows and Mac OS. Platform robustness has been validated using pharmacogenomic data for 23 drugs across four diseases and holds the potential for utilization with any form of tabular data.