Model Research
Tony Ludlow and Liz Atkinson
Model Research is a partnership using computer models and
statistical analyses to solve problems in biological research.
Tony Ludlow was formerly head of Statistics and Computing and head of the
Biometric Modelling Section at the Forestry Commission Research Division,
where he was responsible for developing the Forestry Commission model of the
processes of forest growth.
He did his Ph.D. using computer models to understand inhibitory networks of
neurons and their role in coordinating behaviour. He worked at Imperial College
with John Kennedy F.R.S. in experiments on insect behaviour, and took the lead
in analysing and modelling the results of their joint work. The studies helped to
show how male moths find a female when she is emmitting sex pheromone while
other work focussed on aphid flight and locust walking behaviour. Selected
publications
Liz Atkinson (Ludlow) studied biochemistry while employed in research on
malaria in Liverpool for six years. She worked for three years at Imperial College
on the isoenzymes of the protozoan parasite, Sarcocystis, describing a new species
in her studies before taking a research fellowship at Kings College, London where
she worked on protein metabolism. Selected publications
Model Research has powerful computing equipment running various programming
languages as well as Genstat, GLIM 4, Mathcad, LPA Prolog, DARE P and other
scientific software, together with desk-top and web publishing.
Over 30 years modelling a wide range of biological systems.
A complex system has many interacting processes. The art
of modelling is to find the most important and include them in sufficient detail,
with their interactions. Less important processes must be approximated or left out
altogether.
To make these key choices we must talk to experts and find what is known, why
the model is needed, and what questions it should answer.
We have used models:
- to predict the effect of climate change on forest growth;
- to calculate the optimum spacing of firebreaks;
- to understand the behaviour of networks of neurons;
- to show that a 20-year old theory never did explain what it was
supposed to explain;
- to find gaps in our knowledge when designing experiments;
- to target our statistical analyses
A common thread in these and many other studies has
been that models show the consequences of our assumptions.
- Without a formal model people tend to gloss over missing parts of the
argument and draw unwarranted conclusions.
- The model does not work unless it is complete and the process of
building it shows any gaps and raises new questions.
- The models we build give you the chance to try new ideas: “what if?”
We will meet you to learn about the questions you are asking and
to discuss what is known already. After the preliminary discussion we will write a
framework for the modelling project. The framework:
- Defines the variables that need to be predicted
- Identifies processes which are thought to influence the predicted
variables.
- Gives the model a provisional structure so that we can see more clearly
the types of information that will be needed as inputs to the model.
- Helps to identify appropriate levels of detail and the best resolution in
time and space.
- Allows us to make an estimate of the time and costs, and allows you
to judge the benefits, of the model.
The modelling framework will raise further questions
and will need to be extended to take advantage of expert knowledge.
- Have we left anything crucial out of the framework?
- Are there other questions the model could answer?
- Who will the users be? Are they students learning established theory,
researchers trying to develop new ideas, or managers testing options on
a day to day basis.
- What computing equipment are they used to? What software do they
like?
The data needed for building and testing a model must be
identified and a strategy for testing devised. Data costs and sources need to be
identified and information may come from:
- The client
- A public or commercial source
- A literature review
- New experiments
Whatever the source, it is usually necessary to analyse or transform the data
before it can be used to build or test a model.
Where there are gaps it may still be useful to build a model and try a range of
sensible values to see how much difference they make (sensitivity analysis). How
accurately must gaps be filled?
There should be nothing mysterious about a model. Once the
framework and data sources are agreed, everyone should have a clear
understanding and, in principle, anyone should be able to build it. But we are
probably quicker and we use literate programming tools that show the science not
the ‘do loops’.
A process model can be tested in dozens of ways. Any of its
predicted or intermediate variables can be compared with real life and it could fail
because:
- It predicts the output variables wrongly
- It predicts the right outputs, but needs silly inputs to do so
- Part of its structure is shown to be false by some other experiment.
In contrast, a regression model may only be rejected if it makes the wrong
predictions.
A thorough testing strategy needs to be drawn up as part of the framework and
implemented as the model is being built.
Where a model is being built as part of a research effort it
is inevitable that lessons will be learned and improvements made while it is being
developed. Modifications to the framework will be agreed before changes are made
and further work is undertaken.
We provide documentation at every stage:
- The framework sets out objectives, key variables and the processes
influencing them.
- Programs will be clearly structured with comments so that
non-programmers can check the assumptions and methods.
- The user’s manual will have sections on assumptions, structure, data
sources, test results, running the programs and on how the program
may be extended.
The following lecture notes are available as PDF files and may take some time to
download. The size of each file is given in brackets.
Tony Ludlow has written a number of novel applications in LPA Prolog. Two of
these, Environmental modelling and Identifying fungi in culture, are described on
the LPA web site: www.lpa.co.uk/ind_inf.htm
More recent essays are in bird identification and genealogy: www.ludlowgenealogy.org
To my surprise, I discovered that my paper The behaviour of a model animal
(Ludlow, 1976) is still being cited quite regularly. There are other papers and a
Ph.D. thesis on the ‘Model Animal’. The fullest and most recent account is the
Thesis, so I have scanned it in and made it available as a .pdf file. The 5th
Chapter is a detailed theoretical study of orientation in male moths finding a
female emmitting sex pheromone and proof changes were finally made on 5th
September 2004.
More details and instructions for downloading are available by clicking on:
www.modelresearch.com/science/thesis.
Dr A R Ludlow
Model Research
7 King’s Road
Alton
Hants, GU34 1PZ
Tel/Fax: 01420 83922
E-mail:
info@modelresearch.com
Our core business is statistical
analysis and modelling complex systems, but the techniques we use can be
applied to other projects. All of these services are charged at an hourly
rate which will be negotiated for each contract. Additional charges may
be necessary if we need to buy or rent special software for a particular
project.
Writing a framework usually takes about 2 days.
Discussions before and after the framework document are charged at the normal
hourly rate, plus necessary travel expenses.
The framework document will include an estimate of time needed for each stage of
the main project.
Where a project involves large amounts of data
preparation the charges for data entry will be 75% of the standard rate, subject
to the quality of data sheets.