We implemented and disseminated a stable production version of the FeatureCloud App Store including software development packages for corresponding computer-computer interfaces running as web servers - fostering an advanced, user-extendible App Store functionality. All source code is stored in Git repositories. Several hackathons, including one for external participants, supported app development and platform stress-testing. We have added SMPC and DP functionalities to secure the platform further. More than 50 apps are available now, from standard statistics apps to advanced apps for federated principal component analysis (PCA, see Hartebrodt et al. 2021 and 2022), artificial neural networks, random forest classifiers, and survival time prediction. In total, we prepared eight live demos to illustrate FeatureCloud’s capabilities. Using the first FeatureCloud apps, we demonstrated the power of FML coupled with relevant privacy-enhancing technologies. We worked on typical medical application scenarios, beginning with the federated genome-wide association study (GWAS) tool ‘sPLINK’ (Nasirigerdeh R et al. 2022, published in Genome Biology), which mimics the non-federated standard GWAS tool ‘PLINK’. We demonstrated that (while currently available GWAS software, based on meta-analyses, loses accuracy when data suffers heterogeneously distributed outcomes or confounders) sPLINK gives the same results as the gold-standard tool ‘PLINK’. ‘sPLINK’ implements federated Chi-squared tests as well as federated multimodal linear and logistic regression models. Likewise, we developed the first software for federated survival analysis, namely ‘PARTEA’ (Späth et al. 2022, published in PLoS Digital Health). It combines federated statistical modelling and differential privacy approaches based on Laplacian noise to generate privacy-preserving Kaplan-Meier plots. In addition, we developed, evaluated, and published ‘flimma’, a federated gene expression data analysis tool (Zolotareva et al. 2021, also published in Genome Biology). The FeatureCloud App Store itself was also evaluated and tested in relevant real-world scenarios (Matschinske et al. 2023, published in the Journal of Medical Internet Research). Exploitation and real-world benefits of the project: The FeatureCloud platform and its underlying technology are versatile and can be applied across various markets and use cases. The demand for secure data accessibility and collaboration with external entities, while maintaining confidentiality, is substantial. Consequently, FeatureCloud.ai has been made available as an openly accessible platform. This will serve as a catalyst for app developers, start-up companies, and future projects.
