Multilevel models Improve precision and Speed of IC
            <sub>50</sub>
            Estimates

Vis, Daniël J.; Bombardelli, Lorenzo; Lightfoot, Howard; Iorio, Francesco; Garnett, Mathew J.; Wessels, Lodewyk

doi:10.2217/pgs.16.15

Cited by 69 publications

(72 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Drugs considered in this study 113 had a response in at least 3 cell lines (IC50 lower than half of the maximum screened 114 concentration) and 86% of all possible drug/cell line IC50 measurements have been evaluated 115 ( Supplementary Figure 1a, Supplementary Table 3). Two experimental protocols were used 116 to generate the drug sensitivity measurements, termed here as GDSC1 (Iorio et al, 2016) Figure 1c). For this reason, despite the fact that compounds screened with both technologies 121 showed good agreement (n=66, mean Pearson's R=0.50), we analysed the measurements of 122 the screens separately.…”

mentioning

confidence: 99%

“…Traditionally this has been performed by building predictive models of drug response using 429 mutation, copy number and gene expression (Iorio et al, 2016;Tsherniak et al, 2017). Here 430 we extended this approach, and propose what we term as robust pharmacogenomic 431 association -a drug response and gene fitness pair that are significantly correlated and are 432 also both significantly related to the same molecular biomarker.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Gonçalvès

Segura‐Cabrera

Pacini

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

36Low success rates during drug development are due in part to the difficulty of 37 defining drug mechanism-of-action and molecular markers of therapeutic activity. Here, 38 we integrated 199,219 drug sensitivity measurements for 397 unique anti-cancer drugs 39 and genome-wide CRISPR loss-of-function screens in 484 cell lines to systematically 40 investigate in cellular drug mechanism-of-action. We observed an enrichment for 41 positive associations between drug sensitivity and knockout of their nominal targets, 42 and by leveraging protein-protein networks we identified pathways that mediate drug 43 response. This revealed an unappreciated role of mitochondrial E3 ubiquitin-protein 44 ligase MARCH5 in sensitivity to MCL1 inhibitors. We also estimated drug on-target and 45 off-target activity, informing on specificity, potency and toxicity. Linking drug and gene 46 dependency together with genomic datasets uncovered contexts in which molecular 47 networks when perturbed mediate cancer cell loss-of-fitness, and thereby provide 48 independent and orthogonal evidence of biomarkers for drug development. This study 49 illustrates how integrating cell line drug sensitivity with CRISPR loss-of-function 50 screens can elucidate mechanism-of-action to advance drug development. 51 inhibitors ( Supplementary Figure 1d and1e). 126 127 Cell fitness effects for 16,643 gene knockouts have been measured using genome-128 wide CRISPR-Cas9 screens at the Sanger and Broad Institutes (Meyers et al, 2017; Behan et 129 al, 2019; DepMap, 2019) (Supplementary Table 4). The first PC across the cell lines (6.8% 130 variance explained) separated the two institutes of origin ( Supplementary Figure 2a), 131 consistent with a comparative analysis performed on an overlapping set of cell lines (Dempster 132 et al, 2019). Growth rate was less significantly associated with CRISPR knockout response 133 (Supplementary Figure 2b and 2c). 134 157 are MCL1 and BCL2 selective inhibitors, respectively. Gene fitness log2 fold-changes (FC) are scaled by using 158 Supplementary Figure 2. Overview of the CRISPR-Cas9 datasets. a, PCA analysis of the samples in the 616 CRISPR-Cas9 screens, samples institute of origin is highlighted. b, correlation coefficients between all top 10 PCs 617 and growth rate. c, correlation between cell lines growth rate and PC3 (Pearson correlation coefficient reported in 618 the top left). 619 23 620 Supplementary Figure 3. Drug response and gene fitness associations. a, total number of drugs utilised in the 621 study and the different levels of information available: 'All' represents all the drugs including replicates screened 622 with different technologies (GDSC1 and GDSC2); 'Unique' counts the number of unique drug names; 'Annotated' 623 shows the number of unique drugs with manual annotation of nominal targets; and 'Target tested' represents the 624 number of unique drugs, with target information, for which the target has been knocked-out in the CRISPR-Cas9 625 screens. b, histogram of the drug-gene associations effect sizes ...

show abstract

mentioning

confidence: 99%

mentioning

confidence: 99%

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Gonçalvès

Segura‐Cabrera

Pacini

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Correlating genomic features to drug sensitivity measures such as IC50 values across a panel of cancer cell-lines is a core activity carried out by pharmacogenomics researchers within both academic and industrial settings. There is a lack of literature within this field on the design of dose-response experiments but an interest in exploring new metrics and analytical methods for such experiments (16,18,20,21). The aim of this study was to show how simulation studies can be used to aid in the design of dose-response experiments and assist in the choice of analysis method to be applied.…”

Section: Discussionmentioning

confidence: 99%

“…This approach discards uncertainty in the estimates, and doesn't allow information to be shared between curves. The use of mixed effects models where data is combined across cell lines and drugs has been shown to increase the accuracy of IC50 estimates for large scale screens by sharing information (21). An extension of this approach is to include the genetic covariate itself in the non-linear mixed effects model.…”

Section: Introductionmentioning

confidence: 99%

Simulation of cancer cell line pharmacogenomics data to optimise experimental design and analysis strategy

Mistry¹,

Chapman²

2017

Preprint

View full text Add to dashboard Cite

Background: Explaining the variability in drug sensitivity across a panel of cell lines using genomic information is a key aspect of cancer drug discovery. The results of such analyses may ultimately determine which patients are likely to benefit from a new treatment. There are numerous experimental factors that can influence the outcomes of cell line screening panels such as the number of replicates, number of doses explored etc. Simulation studies can aid in understanding how variability in these experimental factors can affect the statistical power of a given analysis method. In this study dose response data was simulated for a variety of experimental designs and the ability of different methods to retrieve the original simulation parameters was compared. The analysis methods under consideration were a combination of non-linear least squares and ANOVA, conventional approach, versus non-linear mixed effects. Results: Across the simulation studies explored the mixed-effects approach gave similar and in some situations greater statistical power than the conventional approach. In particular the mixed-effects approach gave significantly greater power when there was less information per dose response curve, and when more cell lines screened. More generally the best way to improve statistical power was to screen more cell lines. Conclusions: This study demonstrates the value of simulating data to understand design and analysis choices in the context of cancer drug sensitivity screening. By illustrating the performance of different methods in different situations these results will help researchers in the field generate and analyse data on future preclinical compounds. Ultimately this will benefit patients by ensuring that biomarkers of drug sensitivity have an increased chance of being identified at the preclinical stage.

show abstract

“…For a given cell line in GDSC, the drug response was fitted with a sigmoid curve 55 and consecutively quantified as area under the curve (AUC) or the concentration required to reduce cell viability by half (IC50). GDSC contains 265 compounds tested in 990 cell lines, whilst we focus on a subset of 38 drugs targeting either the PI3K-AKT or MAPK signalling, which leads to 344 experiments considered for evaluation.…”

Section: Pharmacology Datamentioning

confidence: 99%

Defining subpopulations of differential drug response to reveal novel target populations

Toh

Keshava²,

Menden³

et al. 2018

Preprint

View full text Add to dashboard Cite

Personalised medicine has predominantly focused on genetically-altered cancer genes that stratify drug responses, but there is a need to objectively evaluate differential pharmacology patterns at a subpopulation level. Here, we introduce an approach based on unsupervised machine learning to compare the pharmacological response relationships between 344 pairs of cancer therapies. This approach integrated multiple measures of response to identify subpopulations that react differently to inhibitors of the same or different targets to understand mechanisms of resistance and pathway cross-talk. MEK, BRAF, and PI3K inhibitors were shown to be effective as combination therapies for particular BRAF mutant subpopulations. A systematic analysis of the preclinical data for a failed phase III trial of selumetinib combined with docetaxel in lung cancer suggests potential indications in urogenital and colorectal cancers with KRAS mutation. This data-driven study exemplifies a method for stratified medicine to identify novel cancer subpopulations, their genetic biomarkers, and effective drug combinations.

show abstract

Multilevel models Improve precision and Speed of IC ₅₀ Estimates

Abstract: The multilevel model yields a significant reduction of extreme IC50 estimates, an increase in precision and it runs orders of magnitude faster.

Cited by 69 publications

References 8 publications

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Simulation of cancer cell line pharmacogenomics data to optimise experimental design and analysis strategy

Defining subpopulations of differential drug response to reveal novel target populations

Contact Info

Product

Resources

About

Multilevel models Improve precision and Speed of IC 50 Estimates

Abstract: The multilevel model yields a significant reduction of extreme IC50 estimates, an increase in precision and it runs orders of magnitude faster.

Cited by 69 publications

References 8 publications

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Drug mechanism-of-action discovery through the integration of pharmacological and CRISPR screens

Simulation of cancer cell line pharmacogenomics data to optimise experimental design and analysis strategy

Defining subpopulations of differential drug response to reveal novel target populations

Contact Info

Product

Resources

About

Multilevel models Improve precision and Speed of IC ₅₀ Estimates