An automated fitting procedure and software for dose-response curves with multiphasic features

Abstract

In cancer pharmacology (and many other areas), most dose-response curves are satisfactorily described by a classical Hill equation (i.e. 4 parameters logistical). Nevertheless, there are instances where the marked presence of more than one point of inflection, or the presence of combined agonist and antagonist effects, prevents straight-forward modelling of the data via a standard Hill equation. Here we propose a modified model and automated fitting procedure to describe dose-response curves with multiphasic features. The resulting general model enables interpreting each phase of the dose-response as an independent dose-dependent process. We developed an algorithm which automatically generates and ranks dose-response models with varying degrees of multiphasic features. The algorithm was implemented in new freely available Dr Fit software (sourceforge.net/projects/drfit/). We show how our approach is successful in describing dose-response curves with multiphasic features. Additionally, we analysed a large cancer cell viability screen involving 11650 dose-response curves. Based on our algorithm, we found that 28% of cases were better described by a multiphasic model than by the Hill model. We thus provide a robust approach to fit dose-response curves with various degrees of complexity, which, together with the provided software implementation, should enable a wide audience to easily process their own data.

Differential response of human dermal fibroblast subpopulations to visible and near‐infrared light: Potential of photobiomodulation for addressing cutaneous conditions

“…The dose response curve at 450 nm was qualitatively similar to a biphasic dose‐response model with two inflection points (i.e., a classical dose‐response curve models from chemical drug research ), suggesting similarities between the action of light and chemical drugs on cells.…”

“…Standard errors are shown in errobars. (B) Dose‐response curve of human dermal fibroblasts after irradiation with blue light (450 nm) together with the best‐fit (biphasic, two inflection points, best fit: lowest BIC and AIC via Dr Fit software ). Standard deviations are shown in errobars.…”

Coordinated Splicing of Regulatory Detained Introns within Oncogenic Transcripts Creates an Exploitable Vulnerability in Malignant Glioma

“…Fit software package (Di Veroli et al, 2015). Drug activity areas were calculated by summing up the individual differences between a measured activity at a certain drug dose and a fixed reference point of no activity similar to an approach taken in the Cancer Cell Line Encyclopedia (CCLE, (Barretina et al, 2012)).…”

Detecting differential growth of microbial populations with Gaussian process regression

“…Population genome-wide association studies (GWAS) are also a potential application of this method to detect the effect of different loci on function responses (Fusi and Listgarten 2016). By adding new covariates, B-GREAT may also be extended to model continuous effects such as dose response (Sekse et al 2012;Di Veroli et al 2015;Twarog et al 2016). By using a time-dependent variance parameter rather than a stationary kernel, B-GREAT may also be extended to model functional data in which heterogeneity between samples is a function of time (Cao et al 2010).…”