Longitudinal Aspects of in-silico Validation

Mechanisms do not work in a “snapshot mode”. They are active at some stages of the disease and biomarkers (measurable variables) that are part of candidate mechanisms show trajectories. The famous models (plural!) proposed by Clifford Jack abstract from the reality of cohort studies:

The hypothetical standard model for Alzheimer´s Disease has little to do with the reality of biomarkers in clinical cohort studies such as ADNI. Nonetheless, for the identification of patient subgroups, we may need to measure at the right time to have a chance to measure the right variables at the appropriate (informative) activation state of the measurable. In silico validation therefore needs to pay attention to longitudinal aspects and all “static” representation of mechanisms will fall short of discriminating between patient subgroups if we do not take “stages” and “trajectories” into account.

Partner Fraunhofer has developed a “longitudinal study viewer” in the course of IMI project EPAD. The longitudinal viewer currently displays a version of the Clifford Jack models that have underlying numerical data (not just drawings). Furthermore, we have major trajectories of ADNI in that viewer. We will – in the near future – have major trajectories of AIBL in that viewer and populate the longitudinal, “staged” model of AD with more and more relevant, longitudinal information.

Dedicated information extraction approaches allow us, to systematically enrich documents that contain information on the dynamics of biomarkers believed to be relevant for neurodegenerative disease progression. Fraunhofer has established a workflow that finds and retrieves publications with tables that contain relevant clinical study information and has implemented technology that extracts values from these tables and can use that information to simulate variables in patient cohorts. These variables “behave” like the variables in famous and widely recognized studies, such as ADNI. Fraunhofer has also extracted and computed all major trajectories contained in ADNI and is currently importing them into the longitudinal study viewer. This allows for a reality check of abstractions (such like the Clifford Jack Models) and the patient – specific trajectories:

Longitudinal Modeling of Patient Trajectories
Partners UCB and Fraunhofer have developed a methodology to model the longitudinal journey of patients in ADNI and PPMI in a fully multivariate manner, which considers feature dependencies. The model is multi-modal, covers several biological scales and allows for predicting disease severity at each of the observed visits as well as forecasting the effects of interventions (e.g. reduction of alpha-synuclein by 90%). We are currently preparing a corresponding manuscript.

Next steps will be to incorporate the mechanism and data based AD/PD stratification into the framework as well as inclusion of functional brain connectivity. Moreover, we plan to apply the developed method to other cohorts now, including at least one from one of the industrial EFPIA partners.