Barretina 2012 CCL enables predictive modeling of anticancer drug sensitivity

Barrentina 2012 Nature. The CancerCell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity

There are 8 point does-response across 479 cell lines. A logistical sigmoidal function with maximal effect A_max, concentration at half-maximal activity of the compound (EC50), and a Hill coefficient represeting the sigmoidal transition, and the concentration of an absolute inhibition of 50% (IC50). 

947 cell lines were profiled at genomes and expression levels. 

Amazingly, Barrentian12 used the same logistical model with Qin08PONE: 

All dose-response data was reduced to a fitted model using a decision tree

methodology based on the NIH/NCGC assay guidelines

(http://assay.nih.gov/assay/index.php/Table_of_Contents). Models were generated for the

duplicate data points generated for each cell line run day. In brief, dose-response data was

fitted to one of three models depending on the statistical quality of the fits measured

using a Chi-squared test. One approach was the 4 parameter sigmoid model shown

below:

Alternatively, a constant model y = Ainf was employed; or a non-parametric spline

interpolation of the data points was performed (note that this last model represents less

than 5% of models). In these models, A0 and Ainf are the top and bottom asymptotes of the

response; EC50 is the inflection point of the curve; and Hill is the Hill slope, which

describes the steepness of the curve. Other key parameters derived from the models

include the IC50, the concentration where the fitted curve crosses -50%; and Amax, which

is the maximal activity value reached within a model. For the spline interpolation model,

For the spline interpolation model, IC50 and EC50 parameters were both set to the concentration where the fitted model first

crosses -50%. Additionally, we calculated two forms of the Activity area for each curve,

defined as the area between the response curve and a fixed reference Aref = 0 or a variable

reference Aref = max(0, Alow) where Alow is the activity at the lowest concentration, up to

the maximum tested concentration. In practice, the Activity area was calculated as the

sum of differences between the measured Ai at concentration i and the reference level.

Thus, using the fixed reference, Activity area = 0 corresponds to an inactive compound,

and 8 corresponds to a compound which had A = -100% at all eight concentrations

points. The variable reference form was introduced to adjust for curves with large

positive activities close to zero concentration, which are usually artifacts of imperfectly

corrected variations on the assay plate. For this measure, the median of all replicate

activity values was used regardless of cell line run day. To prevent confusion, the Activity

Area was calculated using Aref = 0 unless otherwise noted.