Àlvar Agustí from the Thorax Institute, Barnaclinic+, Barcelona, Spain told delegates to the 2014 Biomax Symposium, held on 5 September at Martinsried near Munich, Bavaria that: ‘These kinds of systems medicine, personalised medicine, P4 medicine, whatever you want to call it, cannot be done by a single type of professional. Clinicians alone cannot do this. Biologists alone cannot do this. Bioinformaticians alone cannot do this. Unless there is multidisciplinarity there is no way we can prevail.’
Personalised medicine requires the integration of data ranging from a patient’s clinical history to their genomic information, as well as pollution and environmental data. This massively increases the computational power needed to process the information, but also makes demands on the intelligence of the software systems that are being used to process the data in order to make the predictions.
Jim Roldan, from Linkcare Health Services, Barcelona, Spain, said: ‘When we talk about big data [in systems medicine], the main reason is because we have added some sources of data such as omics or environmental information’. Roldan went on to say: ‘as you know when you get into this kind of information, the amount of data is enormous.’
Hans-Werner Mewes from the Helmholtz-Center, Munich, Germany, mentioned similar experiences, he said: ‘Over the last years I have been more and more puzzled about how to deal with huge amounts of data and then try to filter out something that was interesting.’
The idea behind all of these systems is to use data and the computational power needed to sort, manage and mine it effectively so that the system can provide knowledge to the users helping them to refine their decision-making. As the systems used for prediction become more complex, it will become more and more challenging to refine the models based on finding the relevant biomarkers and proving the causal link to the disease.
However as Mewes describes: ‘One of the biggest problems in modelling is that almost all the networks are unidirectional and they hardly ever give feedback on the control over time.’ He gave an example of over 1,500 biomarkers of which only a handful had proven useful once they had been sufficiently researched. Mewes said: ‘Most were significant indicators, but they have no predictive power. Predictive power is all about causality.’
To engineer these complex models for use in predictive medicine it is hugely important that the causality of the biomarkers used are fully understood. Once the research has been completed then these kinds of models can be constructed as researchers are trying to do with COPD as mentioned in the first report from the Biomax Symposium.
If computational models for predictive medicine are to overcome their current challenges and transform the treatment of life-threatening conditions, from respiratory disorders to cancer, informatics software will need to be at the heart of data integration. Data analytics will also be required if knowledge is to be derived from the ever-increasing amounts of data.
Roldan said: ‘Knowledge, at the end, is somewhere above the level of data and information. It is about making decisions. When you are a doctor and you have few minutes to visit your patient, you need to understand what to ask this patient, what information you need to get from them, what information you have about them already, so that you can take a decision about the next step to take.’
Roldan finished with ‘There is a trend to be less intuitive and be more effective at matching cost with the effectiveness of the results, in any of these steps Knowledge as a Service (KaaS) allows you to take the right direction or as we say now, gives you the options. What we have learned really is that we cannot give you advice but we can say “according to the knowledge base, these are the possible alternatives”.’
