A statistical framework for assessing pharmacological responses and biomarkers using uncertainty estimates

High-throughput testing of drugs across molecular-characterised cell lines can identify candidate treatments and discover biomarkers. However, the cells’ response to a drug is typically quantified by a summary statistic from a best-fit dose-response curve, whilst neglecting the uncertainty of the curve fit and the potential variability in the raw readouts. Here, we model the experimental variance using Gaussian Processes, and subsequently, leverage uncertainty estimates to identify associated biomarkers with a new Bayesian framework. Applied to in vitro screening data on 265 compounds across 1074 cancer cell lines, our models identified 24 clinically established drug-response biomarkers, and provided evidence for six novel biomarkers by accounting for association with low uncertainty. We validated our uncertainty estimates with an additional drug screen of 26 drugs, 10 cell lines with 8 to 9 replicates. Our method is applicable to any dose-response data without replicates, and improves biomarker discovery for precision medicine.

Inching towards precision medicine for multiple myeloma with causal network models.

e19526 Background: Multiple myeloma (MM) is the 2nd most prevalent blood cancer. For each molecular subtype of MM defined by molecular profiles, mechanisms underlying prognosis and drug response are largely unknown. Elucidating the mechanisms underlying differences in prognosis and drug response will enable a more personalized, precision medicine (PM) approach to treating patients. Methods: Big multiomics data are now widely available and contain the ingredients necessary to build causal models to uncover mechanisms of prognosis and drug response. However, data from resources like the MM Research Consortium (MMRC) dataset may contain errors (eg, sample mislabeling) that diminish modeling accuracy. We developed a probabilistic data matching method ( proMODMatcher) to correct such errors and applied it to the MMRC data. We identified labeling errors in 10 patients, including swaps in RNA and CNV profiles. Given the corrected MMRC data, we characterized recurrent genomic aberrations in MM. We found > 8000 genes significantly associated (FDR < 0.01) with CNVs spanning the gene locations. To distinguish expression correlations driven by causal biological relationships from those driven by coincident CNV influence, we constructed a causal MM network based on the MMRC dataset. Results: Overlaps among multiomic prognostic or drug response biomarkers are sparse. To identify common mechanisms of different biomarkers, we projected them onto our MM network to define the causal context in which they occur. Prognostic biomarkers were enriched in subnetworks associated with mitotic regulation and lipid metabolism, and predicted by our network to be regulated by ZWINT, BUB1B, DTL, TPX2 and NOP16. Proteasome inhibitor and immunomodulating drug responses were mediated by amino acid biosynthesis and cell surface protein complex subnetworks, respectively, with MTHFD2 and TMC8 inferred as the master regulators of these subnetworks, respectively. Regulators include genes (eg, BUB1B and MTHFD2) known to associate with tumor growth and drug response and novel therapeutic control points. Conclusions: Causal models can elucidate the molecular mechanisms underlying prognosis and drug response, enabling the design of more personalized MM treatments.

Fast and robust method for drug response biomarker identification and sample stratification

With unprecedented progress of cancer research, the world is now prepared with versatile arsenal of drugs to combat cancer. However, individual’s response to any drug or combination treatment stands as a major challenge and hence there exists the sheer need for personalized medication. Identification of drug response biomarkers from a wholistic tumor microenvironment analysis would guide researchers to develop custom-tailored treatment regimen.In this study, a fast and robust method has been developed to identify drug response biomarkers from entire transcriptomics data analysis in a data-driven manner. The biomarkers which were identified by the method, were able to stratify patients between responders vs non-responders population. Furthermore, bayesian network (BN) analysis, done on the data, brought forth a mechanistic insight into the role of identified biomarkers in regulating drug’s efficacy.The importance of this work lies with the protocol that is time saving and requires less computation power, yet analyzes a whole system data and helps the researchers to take a step forward towards the development of personalized care in effective cancer treatment.HighlightsSupervised machine learning approach to analyze gene expression data.Drug response biomarker identification.Categorization of samples for their drug response with the help of identified biomarkers.Functional enrichment to understand the biomarkers association with biological processes.Bayesian network analysis to develop causal structure among identified biomarkers and drug targets.Time and cost-effective pipeline for fast and robust prediction of drug response biomarkers.