We established the DC-ren data space, defining kidney relevant process models and embedded biomarker candidates linked with disease progression and drug mechanism of action, and developed two analytic concepts. One focused on prediction of future outcome based on current information (state map approach), while the other followed changes in variables over time that are directly predictive of outcome (state evolution map approach). These efforts were complemented by efforts to make predictive statements by analysis of the urinary proteome. The algorithms approximating the observed variance of eGFR trajectories in the context of treatments of interest (renin angiotensin system blocker only or in combination with SGLT2 inhibitors, mineralocorticoid receptor antagonists or GLP1 agonists) were developed and tested for their sensitivity and specificity in a discovery cohort.
Next DKD mechanisms and their interactions to rationalize the predictive statements for the response to specific treatments were determined. A decision-support toolbox software prototype was developed and a protocol of a virtual clinical validation trial was designed. A systematic literature review was performed to assess the current accuracy of treatment response prediction to be used as a benchmark for the DC-ren results.
The validation trial showed that sensitivity and specificity of the various approaches to predict short term outcome are in line with currently available assessments of long term outcomes, but not superior. Furthermore, it became clear that conceptually both state map and state evolution map algorithms are necessary to capture both organ specific damage programmes and external factors also affecting outcome.
In addition, several application settings were developed (open drug repositioning, cost benefit analysis, patient and expert surveys, design of a clinical trial to test the technical feasability of decision support based clinical routine). Finally the generalizability of the state maps and state evolutions maps in complex diseases beyond DKD and an in-silico workflow for drug response prediction utilizing urinary proteomics profiles were elaborated.
In summary DC-ren has reached its goals but clearly additional work will be necessary to obtain a deterministic tool that can be transferred into routine clinical use. Technically we established two frameworks with associated workflows how conventional staging systems for complex diseases can be replaced by states. Analysis of the results point towards the necessity to combine concepts to capture both, organ specific events as well as interfering external factors. Finally, DC-ren developed a toolbox that is flexible and thus able to host different analysis tools well beyond those that are established within the project run time.
Exploitation beyond project runtime will focus on the following assets:
• an in-silico workflow for drug response prediction utilizing urinary proteomics profiles
• a computational workflow for mechanism-guided time series analysis
• an integrated learning workflow for estimating personalized drug response
• a software toolbox for integrating technically heterogeneous classification functions
• a web-based tool for analysis of ncRNAs in disease phenotypes
For dissemination, project results were presented at national and international meetings and published in the scientific literature, in lay press and media. The DC ren webpage is updated regularly, also holding videos of project partners detailing their contributions. Outreach was further increased via social media presence.
