Paul De Sousa, head of life sciences at Massive Analytic and former researcher at Edinburgh University, writes about a study using artificial precognition AI to analyse results of protein biomarker tests associated with Alzheimer’s disease progression.
Accounting for over 30 million Disability Adjusted Life Years worldwide, Alzheimer’s disease (AD) is a global societal challenge and a threat to healthcare systems around the world. A long history of failures of AD drug trials has highlighted the need for early detection and diagnosis to support patients and clinicians to implement the best life adjustments or medical interventions to alter the course of the disease and personalise the care of those at risk.
Biomarkers are measurable indicators of the biological conditions of health, on which disease prognosis and diagnosis is founded. In AD there are a range of diagnostic procedures to detect these biomarkers including testing Cerebrospinal fluid (CSF) and PET scans for markers of amyloid-β and tau that can accurately detect AD pathology, but their cost and invasive nature preclude the broad accessibility required for early detection. There is thus an urgent need for digital innovations empowering earlier, interpretative and accessible prediction of the risk to develop AD, to allow early access to care and intervention, and streamlining of referrals to specialist centres, which will in turn reduce the burden on healthcare systems.
Applying predictive models in the real world, comes with certain risks due to not fully comprehending errors or bias and randomness, ambiguity, and occurrence of rare events in the data upon which models are built. These all contribute to different kinds of uncertainty. Massive Analytic’s Artificial Precognition (AP) technology combines machine inductive and deductive reasoning to try to overcome data uncertainty hurdles. Within the Oscar Data Science platform, AP provides the user with simplified automated steps to “course tune” the selection of the most pertinent sub-sets of data for further investigation using possibilistic classifiers such as decision trees. Followed by “fine tuning” to arrive at actionable predictions using machine learning algorithms. This process is analogous to human deductive reasoning, meaning the definition and application of rules which makes the outcomes interpretable. This is important to substantiate patient and clinician confidence in predictions. It is also valuable to inform further discovery of factors which may contribute to progression towards or of disease.
In collaboration with Dr Saturnino Luz at the University of Edinburgh Usher Institute, MAL’s OSCAR data science platform and Artificial Precognition were applied to analyse Alzheimer Disease Neuroimaging Initiative blood plasma protein biomarker panel data from subjects followed over time with other tests for brain structure (images) and function (assessments of cognition). ADNI was launched in 2003 as a public-private partnership, led by principal investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment and early AD. The plasma protein data was obtained from “Biomarkers Consortium Plasma Proteomics Project RBM multiplex data,” which contains 190 proteins previously reported in the literature to be related to human pathogenesis.
Working with Steve Cowper, Massive Analytics’ head of data science, the OSCAR platform facilitated analysis of all 190 blood protein features, and achieved 91% sensitivity (correctly detecting positive instances), 92% specificity (correctly detecting negative instances), and a combined 91% predictive accuracy. By comparison a recent study at the University of Tokyo who is also intent on developing an interpretable model of AD diagnosis from the same data, reached a predictive accuracy for AD of 76% when confined to 14 proteins.
Beyond the initial diagnosis, the ability to predict the risk and level of progression from normal cognition (NC) or Mild Cognitive Impairment (MCI) to AD is also a task of interest. Based on the same set of protein features AP was applied to model the risk of progression to AD. We were clearly able to separate classification of those who remained cognitively healthy and those who progressed to advanced AD. Mapping onto these classifications individuals with mild cognitive impairment whose status had not changed over the sample interval, it is possible to identify those at higher risk of progressing to AD in future. Each of these cases has the full suite of biomarker results available against them as well as the two new features generated by the processing. Interpretable tree structures constructed from these features can be easily generated.
Our pilot study demonstrated the power of the OSCAR platform and AP to yield new actionable insights in AD diagnosis and prognosis. These are now being extended to integrate other clinical and non-clinical parameters and to incorporate human/AI hybrid decision making in clinical decision support and medical devices.
