SP3 - COMMITMENT

Systems medicine knowledge and mechanisms

Successful translation of systems medicine approaches to application in psychiatry will depend on the ability to accurately model pathogenetic mechanisms and make these available for machine learning analysis. To address this, SP3 subproject will identify and extract mechanistic information from the scientific literature and capture expert knowledge from the participating biomedical and clinical research partners. We will represent this a-priori knowledge in computable “cause-and-effect” models based on OpenBEL, the Biological Expression Language. Encoding a-priori knowledge in computable graph models (so-called “knowledge assemblies”) allows to establish entire inventories of mechanism graphs specific for a clinical indication area. We build on substantial experience in generating such mechanism-inventories in the area of neurodegenerative diseases. We have previously demonstrated that mechanism-inventories can be applied on patient-level clinical data sets for mechanism-based stratification of patients. SP3 – knowledge and mechanisms will make the vast majority of relevant knowledge accessible to transfer learning, effectively supporting the inclusion of a-priori knowledge during machine learning approaches developed in SP4 and applied in SP5 and SP6. Dedicated knowledge assemblies will be generated that support co-morbidity modelling by focusing on “shared pathways” (shared and crosstalking pathophysiology mechanisms). Subproject 3 has two work packages: WP1 focuses on technology development that enables us to generate the inventory of mechanisms underlying psychotic illnesses and comorbid conditions. This will allow machine learning algorithms developed in SP4 and applied in SP5 and SP6 to be tuned towards mechanistically-important information. In WP1, we will further develop the technologies and workflows for maintaining, extending and updating this highly specific knowledge resource. Also included in this work package are all technology developments required to integrate the resource in the information technology (IT) infrastructure (SP1) and for knowledge service provisioning. In WP2 we apply the knowledge resource to optimize the utility of mechanism information for machine learning applied in COMMITMENT (SP5 and SP6). Following expert curation of the knowledge assemblies, we focus on application scenarios and algorithms for their optimal use. In particular, strategies for the inclusion of graph-based knowledge representations in machine learning and transfer-learning approaches will be developed in this work package.