Mark of time

A pioneering study at Manchester University is using a ‘robot scientist’ to examine blood samples for biological markers that may diagnose Alzheimer’s disease.

‘Unlike diabetes or blood pressure, there is no simple diagnostic test for Alzheimer’s disease,’ said Dr Nitin Purandare, a senior lecturer and honorary consultant in old age psychiatry at Manchester.

The robot scientist combines the automatic operation of a blood analysis technique called GCGC-MS with artificial intelligence to determine which experiment to carry out next.

It will analyse blood samples from 500 people with diagnosed Alzheimer’s disease and a control group of 500 people, comparing levels of thousands of different metabolites. The goal is to discover diagnostic and prognostic biomarkers, which may point the way to a treatment.

‘But the first step is simply to identify which chemicals in the blood separate Alzheimer’s patients from the control group,’ said Purandare.



Diagnosing dementia

According to The Alzheimer’s Society, dementia affects one person in 20 over the age of 65, and one person in five over the age of 80. The number of people with dementia is steadily increasing.

Alzheimer’s disease is the most common form of dementia, making up 55 per cent of all cases of dementia. It is a devastating, progressive condition that causes physical damage to more parts of the brain as it advances. Although associated with a shortage of certain chemicals in the brain, it is difficult to diagnose, with assessment carried out using brain scans, interviews and memory tests. True confirmation is only possible post-mortem.

GCGC-MS is two phases of gas chromatography followed by mass spectography to identify substances in each blood sample. The second gas chromatography phase gives better separation, allowing some of the identifying peaks that would otherwise be hidden under others to be revealed.

‘The robot will try to match different sets of chemicals, said Purandare. ‘It could, for example, compare levels of a set of 100 chemicals in patients against levels in the control, and if there were no pattern, it would try again, learning from its own mistakes. At the end we hope to have a group of chemicals the levels of which will be a “signature” of Alzheimer’s.’

The next phase will examine individual symptoms such as depression, losing interest and hallucinations, using the same approach to further break down which chemicals are associated with each symptom. There will be a follow-up after one year to see if any of the control group with prognostic levels of the biomarkers displays early symptoms such as memory loss.

Douglas Kell, a professor of bioanalytical science at Manchester, was one of the developers of the robot scientist.

‘The original idea was to automate the process of scientific discovery,’ said Kell.

‘There is a model by which we alternate the world of ideas with the world of experience. We carry out an experiment then revise our hypothesis in a cyclic loop. The robot scientist can combine working out what experiment is best to do next with actually carrying it out.’



Closed-loop learning

The robot uses Inductive Logic Programming, a machine learning process. The scientists give it the background knowledge about the experiment, called the domain. It then decides which hypothesis to follow using the available data.

It uses multi-objective optimisation to decide what experiments to carry out to prove the hypothesis and carries them out with no human intervention. This process is called closed-loop machine learning.

‘Typically, you have a load of data and carry out data mining,’ said Kell. ‘In active learning, you have a subset of the data. You know the next data point is at a certain point. It can learn the most efficient path between the two.’

The robot not only operates the scientific instruments, but also automatically improves the way they operate.

‘You can tweak parameters in the machine software. But if there are 20 parameters with 10 values each, there are 20 to the factor 10 combinations,’ said Kell.

‘You do an experiment and, based on its results, decide which to do next, changing the parameters each time. We automated that, writing software that works on top of the machine software. We have seen three to four times more peaks using this method.’

The ultimate goal is to discover diagnostic and prognostic biomarkers that can give indicators for treatment. ‘We hope we don’t stumble on side-markers,’ said Kell. ‘We need to discriminate between causation and correlation.’

The robot scientist has already proven successful. ‘In our pre-eclampsia studies we found three novel biomarkers,’ said Kell.