软件技术如何帮助开发更精确有效的体外诊断系统?
The
use of in vitro tests often plays an important role in the diagnosis
and treatment of disease. However, many of these tests rely on detecting
single biomarkers, which does not always provide a complete picture.
We
spoke to Dr Philipp Pagel, CMO at numares AG, to learn about some of
the limitations of current in vitro diagnostics, and how software based
technologies can help overcome some of these issues.
Anna MacDonald (AM): Can you tell us about some of the limitations of current in vitrodiagnostics?
Philipp Pagel (PP): Current
laboratory work is focused on individual parameters that are
interpreted as indicators of certain conditions. E.g. if your C-reactive
protein (CRP) goes up that's a sign of inflammation. Lab results are
great because they are quantitative and apparently objective. So for
decades, both academia and industry have put a lot of effort into
discovering new diagnostic biomarkers hoping to cover each and every
disease. However, while there is no shortage of molecules in the human
body to study and plenty of candidates have been proposed, the number of
successfully introduced novel tests is depressingly low. Apparently, it
has become increasingly difficult to develop successful single marker
tests that really deliver on their promise. I believe that the reason
for this may well be that most of the good single-molecule markers have
been found already. Diseases often have complex dynamics and many things
happen in the organism - signal transduction, enzyme activity,
dys-regulation of biochemical pathways and shifts in physiological
states. More often than not hundreds of things are affected making it
unlikely that each and every disease actually has a single marker that
is both sensitive and specific to that particular condition. I.e. many
markers may increase or decrease in concentration in response, but not
exclusively in one disease. Additional information is needed to
distinguish between diseases.
AM: How can software based technologies help overcome some of these limitations?
PP: Although
good single markers may not exist for many diseases that does not
necessarily mean that we cannot distinguish them - but we have to look
at several pieces of information at the same time. I.e. instead of
measuring one marker we can measure e.g. 5 different ones and then let
software combine this information for better sensitivity and
specificity. The individual markers may not have a very good predictive
power, but when combined in the right way the relations between them
indicate the disease. We call these metabolic constellations: If I were
to show you pictures of the stars Betelgeuse, Rigel, Bellatrix, etc. you
will probably not be able to find them in the sky – if I show you a
picture of Orion – you’ll find it immediately. It’s the characteristic
constellation that carries the useful information.
A
chemical assay can only measure one thing – but if your assay is
software that opens an entirely new dimension. Now you can analyze many
metabolites from a single measurement, or re-analyze a previous
measurement with a different question without going back to a stored
sample. You are even able to run tests on that stored data that weren’t
even developed when the measurement was carried out. This is certainly
useful for study cohorts but also when looking into the history of a
current patient.
AM: Can you tell us more about how numares is combining NMR technology with AI?
PP: Those
two technologies are a great match! NMR allows us to quantify many
metabolites at once in a single measurement. We do not need to know in
advance what we are looking for because NMR simply returns a spectrum
that we can then analyze. The technology has a great dynamic range, is
very precise and allows very simple sample preparation. The last point
is important because typically, much of the imprecision found in an
assay comes from pre-analytic steps. Artificial intelligence is our
development tool of choice: These methods are able to detect the
metabolic constellations that indicate health vs. disease or different
types of a disease. In order to do so, we put together large data sets
of relevant clinical information obtained in studies on one hand and
hundreds of metabolites per patient samples measured by NMR on the other
hand. The AI then identifies suitable makers as well as the best way to
interpret them as a metabolic constellation. Of course, that does not
mean that the computer does all the work and no human expertise is
required – before the computer can do its magic, it takes a lot of
expert work to get the right data, clean it, look for abnormalities,
integrate it, understand the biochemistry and pathophysiology of the
constellations found etc.
AM: What advantages has this brought to diagnostics?
PP: I
think, our approach has the potential to tackle diagnostic questions
that are not amenable to single biomarker diagnostics. As mentioned
above our tests consist of software rather than antibodies or chemistry
so the development process is rapid and we should be able to put
together a sizable test menu in a very short period of time giving our
customers a new platform with a growing set of test options. Although
the instrument is a major investment, we are able to offer the tests for
a very competitive price. Unlike many other innovative diagnostic
technologies, we do not require price points in the many hundred dollar
range or above so market adoption should go much smoother. We believe
that we have mastered the technological challenges well at this point so
we can focus on growing the menu. I think that numares' NMR-diagnostics
will quickly become a staple in the major labs.
Philipp Pagel was speaking to Anna MacDonald, Editor for Technology Networks.