Improved large-scale prediction of growth inhibition patterns using the NCI60 cancer cell line panel

Abstract

Motivation: Recent large-scale omics initiatives have catalogued the somatic alterations of cancer cell line panels along with their pharmacological response to hundreds of compounds. In this study, we have explored these data to advance computational approaches that enable more effective and targeted use of current and future anticancer therapeutics.Results: We modelled the 50% growth inhibition bioassay end-point (GI50) of 17 142 compounds screened against 59 cancer cell lines from the NCI60 panel (941 831 data-points, matrix 93.08% complete) by integrating the chemical and biological (cell line) information. We determine that the protein, gene transcript and miRNA abundance provide the highest predictive signal when modelling the GI50 endpoint, which significantly outperformed the DNA copy-number variation or exome sequencing data (Tukey’s Honestly Significant Difference, P <0.05). We demonstrate that, within the limits of the data, our approach exhibits the ability to both interpolate and extrapolate compound bioactivities to new cell lines and tissues and, although to a lesser extent, to dissimilar compounds. Moreover, our approach outperforms previous models generated on the GDSC dataset. Finally, we determine that in the cases investigated in more detail, the predicted drug-pathway associations and growth inhibition patterns are mostly consistent with the experimental data, which also suggests the possibility of identifying genomic markers of drug sensitivity for novel compounds on novel cell lines.Contact: terez@pasteur.fr; ab454@ac.cam.ukSupplementary information: Supplementary data are available at Bioinformatics online.

How Consistent are Publicly Reported Cytotoxicity Data? Large‐Scale Statistical Analysis of the Concordance of Public Independent Cytotoxicity Measurements

“…[30] and Kalliokoski et al [31] set the basis to estimate the maximuma chievable performance of in silico models trained on data issued from different laboratories, which is quantified through the maximum achievable R 2 values on atest or external set. [39] While cellular sensitivity data are of relevance in predictive modeling as aw hole, [33][34][35][40][41][42] and in the fields of cell-line sensitivity and toxicology modeling in particular, [40,41,[43][44][45][46][47] to date, no systematic study has evaluated the comparability of public in vitro cytotoxicity data on al arge scale. To address this shortage, we have implemented in Rapipeline for the automatic extraction and curation of cell-line sensitivity data from ChEMBL version 19.…”

Elucidating Compound Mechanism of Action and Predicting Cytotoxicity Using Machine Learning Approaches, Taking Prediction Confidence into Account

“…It is important to consider that correlation metrics depend on the range of the dependent variable, and hence one might obtain low errors in prediction (i.e., low RMSE values) and yet a low R 2 value if the dependent variable spans few bioactivity units (Alexander, Tropsha, & Winkler, 2015;Cortés-Ciriano et al, 2016). Determining whether a given model shows good generalization capabilities depends on the drug discovery stage in which it is applied.…”

Machine learning models to predictin vivodrug response via optimal dimensionality reduction of tumour molecular profiles

“…We used the recommended values for RF hyperparameters (1000 for the number of trees and the square root of the number of considered features for m try ). We preferred this to tuning these hyperparameters for each training set, as RF tuning generally results in just marginal improvements at the cost of being much more computationally expensive 38,56,57 . As no model selection was carried out for this algorithm, standard LOOCV was performed to estimate the performance of RF using all the features (RF-all) on each data set (treatment-cancer type-molecular profile).…”

“…Indeed, while typically only tens of tumours have their response to the drug available, the molecular profiles of these tumours may easily aggregate over 50,000 features. To face this challenge, ML algorithms with built-in FS such as Elastic Nets 32-36 , Ridge 33,36 , LASSO 33,34,36,37 or Random Forest (RF) 12,33,35,36,38,39 have been used to model pharmacogenomics data from in vitro cell lines. For instance, RF ignores those features irrelevant for predicting drug response and thus has been able to tackle to some extent this challenge.…”