Cancer therapies are often associated with a ‘Goldilocks Window’ of optimal dosing: striking a balance where efficacy is maximised and toxicity is minimised. While mathematical modelling offers a powerful tool to inform optimal treatment approaches in cancer drug development, determining the right level of mechanistic detail to include in a model also requires careful consideration – hitting the right balance is both art and science.
You say mechanistic, I say empirical …
Mechanistic models try to reflect some of the underlying biological processes that govern cancer and hence tend to be more complex. On the other hand, empirical models represent the relationship between the input and output variables based mainly or solely on observed data.
In essence, an empirical model can help you determine the how, while a mechanistic model captures both how and attempts to provide insight into why. Both have a role to play.
What was the question again? …
It sounds obvious but you should always begin with the end in mind. What is the question you’re trying to answer? Do you want a predictive tool, or do you want a framework for exploring the properties of a biological system in a more open-ended way? This will then guide what data you need and what type of modelling approach may best fit the bill.
When less is more…
When it comes to anticancer therapies, complex mechanisms of action and the sheer number of drug combinations and regimens can be daunting. Modelling can help, but the key is to identify the elements of the system that are meaningful and relevant to the question being posed.
In general, if a simple empirical model does the job you want it to do, then consider using it! One common mistake is to think that you must build all the information available to you into your model. No!! If a reasonable simplification of the system enables you to build a model that meets your requirements, then rest… your work is done.
When more mechanistic detail is needed to reproduce data of interest, or to dig further into a drug’s mechanism of action, then remember:
- Research what is known about the biology of the system you’re modelling;
- Don’t be too ambitious; start simple and build upwards;
- Try and stick to including features that you know will be quantifiable, based on the availability of data;
- Don’t overparameterize/ over complexify!
“Often it turns out that less is more. This is based on our experience that sometimes too much detail is counterproductive, particularly when you don’t have the data. Any mathematical model is only as good as the data on which it is based. Otherwise, you’re just introducing more assumptions, more guesses, which doesn’t necessarily help to make a good prediction.” Dr Frances Brightman, Project Leader
Examples from our experience…
Physiomics can build models from scratch or utilise its in-house modelling platform called Virtual Tumour (VT) which is semi-mechanistic. Over the years we have built add-on modules for VT that have enabled us to accurately reproduce multiple data sets generated by different authors in different labs. In our immuno-oncology module, for example, we’ve simplified many elements of the biology, while capturing key events at the heart of cancer immunotherapy, such as the activation of T cells.
“Choosing the right level of complexity means drug developers can more effectively guide drug development and avoid costly and time-consuming mistakes. Ultimately, your choice of model goes back to the question you’re trying to answer.” Dr Frances Brightman, Project Leader, Physiomics
Another example where a semi-mechanistic approach is useful is to explore combinations of drugs that target different parts of the same pathway. For example, agents that target the DNA damage response (DDR). Here we’ve developed a module that represents different types of DNA damage, including single-strand and double-strand breaks, and includes different DNA repair pathways. This allows us to represent the mechanisms of action of DDR agents at a sufficient level of detail to capture the synergies between them and explore their combination effects.
When it’s an art you need artists…
Determining the right level of mechanistic detail to include in a model requires experience specific to the therapy area in question. At Physiomics, we’ve completed over 100 projects, each one tailored to the needs of our clients. Interested in learning more? Get in touch with our team today.