Patients with similar pathologies frequently respond differently to treatments. Precision medicine aims to identify treatments that are more likely to be effective in individual patients. To this end, several studies have been investigating the correlation between drugs sensitivities and gene expression patterns or mutations. However, results vary from study to study and hence are not reliable.
A new algorithm, called MERGE, was developed by researchers from University of Washington, including Associate Professor Su-In Lee, PhD and graduate student Safiye Celik from the Paul G. Allen School of Computer Science and Engineering, Sage Bionetworks and Fred Hutch researcher Dr. Pamela Becker (Clinical Research Division) to predict how tumor cells from individual patients would respond to chemotherapeutic drugs. MERGE integrates more parameters contributing to cancer progression than previous algorithms used to predict drug responsiveness. The results of this study were published in Nature Communications.
While most algorithms generally focus on one parameter, MERGE includes five driver events of cancer progression: gene expression, mutations, copy number, methylation and known regulators of gene expression. These data were obtained from publicly available databases. MERGE ranked the importance of these events for each gene. The genes were then attributed a priority score called the MERGE score indicating the likeliness to be positively or negatively associated with drug response. Gene expression was the most important contributor to the MERGE score attribution while methylation was the least important contributing factor.
To validate MERGE, genome wide expression and drug sensitivity of 14 Acute Myeloid Leukemia (AML) cell lines and cells from 30 AML patients were assessed in vitro against 160 chemotherapeutic drugs. Among the tested drugs, 62 were FDA-approved. Fifty-three drugs were further selected because of their efficacy on at least half of the samples. These drugs presented overlapping mechanisms of action classified into 24 groups. A gene should be comparably sensitive to different drugs with similar mechanisms of action. Such drug class specificity is important for validation of the genes identified as molecular markers of drug sensitivities.
The most statistically significant gene-drug pair associations that also present drug class specificity were selected, allowing the establishment of a list of the top 8 genes. Expression of FLT3 (FMS like tyrosine kinase 3), CASP8A2 (caspase 8 associated protein 2), L2HGDH (L-2-hydroxyglutarate dehydrogenase), MNT (MAX Network Transcriptional Repressor), BAZ2B (Bromodomain Adjacent to Zinc Finger Domain 2B), MZF1 (Myeloid Zinc Finger 1), BEX2 (Brain Expressed X-Linked 2) and SMARCA4 (SWI/SNF Related, Matrix Associated, Actin Dependent Regulator Of Chromatin, Subfamily A, Member 4) genes modulated the sensitivity of cancer cells to various class of drugs such as BCL-2, histone deacetylase or topoisomerase II inhibitors.
The genes identified as predictive molecular markers of sensitivity to chemotherapeutic drugs by MERGE were validated in in vitro experiments. SMARCA4, for example, was predicted to increase cellular sensitivity to topoisomerase II inhibitors. In agreement, AML cell lines became more sensitive to etoposide and mitoxantrone upon overexpression of SMARCA4 protein following transfection.
When asked about future development, Dr. Becker explained, “We hope to soon apply these correlations to choice of drugs for patients in clinical trials; for example, if a patient has high expression of one of these genes, the corresponding drugs or drug classes could be considered or avoided, based on prediction of drug sensitivity or resistance. As this study was performed based on patient samples with AML, similar machine learning algorithms could be developed for other cancers.”
This study was funded by the National Institutes of Health, National Cancer Institute of the National Institutes of Health, National Science Foundation, American Cancer Society, Life Sciences Discovery Fund and philanthropic funding from Norman Metcalfe.
Research reported in the publication is a collaboration between Cancer Consortium members Estey Elihu and Vivian Oehler (Fred Hutch), Pamela Becker, Anthony Blau and Chris Miller (UW).
Lee S-I,Celik S,Logsdon BA,Lundberg SM,Martins TJ,Oehler VG,Estey EH,Miller CP,Chien S,Dai J,Saxena A,Blau CA,Becker PS. 2018. A machine learning approach to integrate big data for precision medicine in acute myeloid leukemia. Nature Communications. 9(1), 42.