The long-term goal of the VITA project is to develop an expert system that can predict the most effective treatment for tumours that express particular biomarker profiles. The approach will be validated initially by predicting optimal treatment for tumours of defined biomarker profiles in mice.
VITA is an inter-connected system of two databases, a rule-based expert system, and a PK/PD model of tumour growth and treatment (Fig. 1). It permits simulation of the treatment of tumours with drugs, surgery, and radiation. The Automated Protocol Generator (APG, module A) is a rule-based expert system that surveys all possible treatment protocols for a particular tumour, and assigns priority scores to them based upon preclinical data, clinical activity, biomarker expression, known drug-drug interactions, and cross-resistance patterns. The biomarker database (module B) is a repository of biomarker measurements in experimental or clinical tumour samples, along with known treatment outcomes for tumours of defined expression patterns. It produces two kinds of output: (i) the biomarker expression pattern for the tumour to be treated is fed to module A and used for prioritisation of potential treatment protocols. (ii) A file of biomarker and cytokinetic parameter data is generated for the tumour to be treated, and fed to module C. The core PK/PD module (C) is a cytokinetic description of the cell cycle, linked to one or more signalling pathways, a model of apoptosis, and several other biological properties. One or two normal cell populations can be modelled simultaneously, making it possible to model drug effects on normal tissues, such as bone marrow and gastrointestinal epithelium. In principle, other normal tissues populations that are sites of drug toxicity (e.g. skin) can be added. Additional toxicities that can be modelled in outline include immunosuppression, cardiotoxicity, and mutagenesis. Tumour properties are described as a list of cytokinetic parameters and biomarkers. Module C requires three forms of input: (i) a tumour descriptor file, of biomarker and cytokinetic data from the biomarker database; (ii) drug descriptor files for each of the drugs in the protocol, containing PK and PD data. This information is acquired indirectly, via the APG. (iii) a protocol file, describing doses and timings for the drugs in the current protocol. The output of module C is a range of measures of antitumour activity and normal tissue toxicity which may be considered as recommendations for actual treatment protocols. Module D is a drug database, containing values of PK and PD parameters for clinically used anticancer drugs and experimental agents.
The model is interactive at three levels: Predicted treatment results generated by module C are fed back to the APG, which may use this data to modify future priority estimates by updating its weighting factors. Actual treatment outcomes (PK, PD and clinical endpoints) are fed back directly to the biomarker database, which uses them to update its correlation of biomarker expression with treatment response. This information is thus indirectly available to the APG. In this way, VITA is able to learn from both predicted and clinical results. In addition, the parameter values of the PD model will be revised as necessary to bring its predictions into line with observation.
Figure 1. Major components of VITA
Future development of VITA will extend the concept of optimising treatment based upon biomarker expression to predicting the most effective treatments for human cancer. A version of VITA is under development that uses human tumour biomarker measurements and PK/PD of anticancer drugs in humans. This expert system for designing personalised medicine for human cancer is termed the Virtual Interactive Patient™, or VIP. A preliminary version of VIP is available to research collaborators, and it is hoped that this complex interactive system will prove a useful tool in the process of matching effective treatments to the highly complex and diverse gamut of human cancers.