Efficient parameter estimation for agent-based models of tumour growth

Overview

Cancer is an extremely heterogeneous disease, particularly at the cellular level. Cells within a single cancerous tumour undergo vastly different rates of proliferation based on their location and specific genetic mutations. Capturing this stochasticity in cell behaviour and its effect on tumour growth is challenging with a deterministic system, e.g. ordinary differential equations, however, is possible with an agent-based model (ABM). In an ABM, cells are modelled as individual agents that have a probability of proliferation and movement in each time-step. This form of model allows us to capture the spatial evolution of cells within a tumour as well as the stochastic evolution of tumour growth.

Estimating the parameters of such ABMs based on data is a difficult task. Standard estimation of parameter estimation are not available due to the unavailability of the likelihood function of the ABM. Thus, we resort to so-called likelihood-free methods that perform parameter estimation by simulating data from the model and seeing if it is close to the observed data. However, these ABMs can be computationally expensive to simulate, rendering many likelihood-free approaches infeasible.

To develop an efficient parameter estimation approach for complex ABMs of tumour growth, this project will explore cutting-edge likelihood-free methods from machine learning that require significantly fewer model simulations.

Research activities

Research activities include:

• implementing likelihood-free methods for complex ABMs of tumour growth in Python
• running the methods on simulated data to investigate their performance
• running the methods on real data
• writing up the results.

Outcomes

The primary outcome of this project is the implementation of cutting-edge likelihood-free methods for performing parameter estimation of complex ABMs of tumour growth.

If done as a VRES, the project would naturally extend into an Honours or MPhil program of research.

Skills and experience

It is desirable but not necessary that you have the following skills:

• proficiency in Python or a similar programming language
• have an understanding of statistical inference (e.g. MXB341).

Keywords

• statistical inference
• stochastic modelling
• bayesian statistics
• data science
• agent based models
• mathematical modelling

Contact

Contact the supervisor for more information.