Development of PK/PD Model Platforms to Support the Optimal Deployment of New Drug Combinations for the Treatment of Malaria

In 2016, there were an estimated 216 million malaria cases and 445,000 deaths from the disease worldwide. While many therapies to the disease currently exist, resistance to many of these treatments are on the rise. It is becoming increasingly important to have accurate predictions about the potential clinical activity of newly proposed dosing regimens (especially those utilising drug combinations) before testing them in clinical studies. Such accurate predictions will help save resources and accelerate the drug development process.
While the prediction of the activity of single drugs in malaria has been fairly successful in the past, the prediction of the overall activity of drug combinations against malaria is extremely more complicated and has not been equally successful. Different anti-malarial drugs act on different stages of the parasite life cycle; this introduces a level of complexity that makes current standard PKPD models less likely to accurately predict the overall clinical activity of different drug combinations. Additionally, a large number of drug activity assays exist for anti-malarial compounds; these include in-vitro assays with different pharmacological outputs as well as in-vivo assays where the drug is tested in infected animals with or without an active immune system.
The activities proposed in this project are expected to result in developing a new mathematical framework that will consolidate the complexity of data derived from different drug assays that have been performed with anti-malarial compounds. This would result in simultaneously translating diverse outputs from different labs into tangible predictions about potential clinical activity of drug combinations. State of the art mathematical modelling will be used to address the issue (e.g. machine learning and artificial neural networks).
The predictions generated using this mathematical framework will be validated against results from clinical studies performed on the field. If the model was successful in predicting clinical activity then it will become a powerful tool that can select for new drug combinations that can achieve maximal activity on the field.
Ultimately, this mathematical tool will have the power to assess the potential of different drug combinations that are currently in use and combinations proposed for clinical studies. This will help with decision making in clinical trials and will have the potential of altering the policy in which such combinations are applied in the field. The predictions will further assess the overall exposure of drug combinations to assess the potential of development of resistance.

The mathematical prediction of drug combination activity against malaria remains to be a challenging area of research. While previous models have been successful in generating meaningful predictions from pre-clinical single drug activity, the prediction of clinical outcomes from drug combination activity in-vitro and in-vivo has resulted in limited success. This is due to the complex nature of the malaria life cycle and the various mechanisms of action of various drugs within different parts of this life cycle. The variety of disease models used to assess the activity of drugs and drug combinations adds an extra layer of complexity where current standard models are unable to consolidate data of different classes (e.g. metabolic activity in in-vitro and parasite reduction rates in an in-vivo assay).
We propose here to develop a novel model that has the ability to consolidate drug activity data from different disease models and that can interpret the drugs' activity within the context of their mechanism of action within the parasite's life cycle. This is categorically different from using standard PKPD combination methods (e.g. Loewe, Bliss or Gaddum non-interaction reference, which are methods that don't account for the presence of several sub populations with different susceptibility to drug action). Additionally, we will use advanced methods of machine learning and artificial neural networks to consolidate the complexity of drug combination activity within different disease models. This will allow for using larger amounts of data to inform clinical predictions rather than being limited to one source of data such as the SCID mouse model.
The predictions generated from this model would then be validated against existing clinical data and would be used to generate new drug combination regimens that can potentially be tested in the malaria human challenge model or in naturally infected patients in the future.

A large number of malaria disease models currently exist to predict the clinical efficacy of anti-malarial compounds. While these models have been fairly successful in ranking different anti-malarial compounds, a large gap still exists in predicting the efficacy of combination anti-malarial therapies. This is because of the complexity of such combinations, the fact that different drugs act on different parts of the parasite life cycle and disparity of different disease models. There is an urgent need to develop mathematical models that can consolidate the complexity of these data to accommodate for the large number of drug combinations that are currently under development to translate their preclinical findings into tangible clinical predictions. The lack of such mathematical models has been confirmed by modellers at Medicines for Malaria Venture (MMV).
The activities that will be undertaken throughout the time of this fellowship aim at delivering a much-needed mathematical tool for predicting the activity of drug combinations against malaria. Such a novel tool will result in predicting the activity of drug combinations that are applied in clinical studies and can be used to advise on appropriate dosing and treatment durations. This will allow for a more rational drug design for clinical studies implementing drug combinations in malaria as current studies are not based upon rigorous predictions of potential activity. Poorly designed dosing regimens that are not driven by rational PKPD modelling could potentially result in unfavourable outcomes of clinical studies in infected patients or in subjects with controlled infection using the human challenge model.
In addition to aiding the design of new clinical trials, the models developed in this project will evaluate currently used drug combinations and could advise on ways to improve their application by altering dosing, treatment duration or choice of drugs. This would allow for safer application of certain anti-malarial therapeutic agents as well as for reducing chances of sub-optimal drug exposure which could ultimately result in the development of parasite resistance to drugs.
We will use a number of mathematical modelling approaches to identify the best possible predictor of drug combinations against malaria. These tools will include machine learning and artificial neural networks which are currently emerging as new tools to predict the PK-PD relationship of different chemotherapeutic agents. The successful utilisation of such tools for malaria can have out reaching applications that would involve the prediction of combination therapies for other infectious or non-infectious diseases. This would ultimately expand the impact of this project to areas beyond malaria.