Jump to accessibility statement Jump to content

SoCoBio (Universities of Southampton, Kent, Sussex, Portsmouth and NIAB EMR)

Prediction of drug-induced renal injury using machine learning applied to in vitro and in silico parameters (CASE project)

Primary Supervisor: Taravat Ghafourian (University of Sussex)
Alex A Freitas (University of Kent)
Neil Crickmore (University of Sussex)
Paul Walker  (Cyprotex)

Project Summary

The kidney’s role as a major route of clearance of xenobiotics and its ability to concentrate the glomerular filtrate make it particularly vulnerable to drug-induced renal injury (DIRI). Meanwhile, due to species-specific differences, animal experiments are unsatisfactory predictors of human nephrotoxicity. Moreover, in-vitro assays using various renal cells often lack the expression/function of transporters and other functional proteins implicated in nephrotoxicity.
Here, we propose to develop a hybrid model for DIRI prediction using machine learning applied to a wealth of in-vitro and in-silico parameters including measures of mitochondrial function, various transporter affinities, molecular descriptors and structural alerts. Our laboratories present a unique opportunity for this multidisciplinary research since we have established collaborations between molecular modelling/cheminformatics lab (University of Sussex) and machine learning lab (University of Kent) as well as industrial support for high-throughput in-vitro experimentations (Cyprotex). Cyprotex is an ADME/Tox company (part of Evotec) with whom we have established agreements, currently hosting one PhD student’s industrial placement. Cyprotex will provide the high-throughput in vitro facilities and assay protocols, including Seahorse XF96 analyzer and high-content imaging assays.
The objectives of this research are:
1) Obtain bioenergetics measures from mitochondrial stress tests, e.g. oxygen consumption rate and extracellular acidification rate.
2) Obtain drugs’ effect on cellular health measures from High Content Imaging, using primary RPTEC (Renal Proximal Tubule Cells).
3) Obtain molecular descriptors and structural fragments from molecular modelling and QSAR software.
4) Develop computer-based models (Quantitative Structure-Activity Relationships, QSAR) for the estimation of renal transporter binding, including MATE, OAT, OCT, URAT, OATP and P-glycoprotein. This is necessary since the cell-based measures lack these functions which have a crucial role in DIRI.
5) Use advanced non-linear classification algorithms from machine learning to develop a model comprising parameters obtained in objectives 1-4, for the classification of DIRI and non-DIRI compounds, and evaluate the reliability of the proposed model by statistical external validation and applicability domain analysis.