The FeatureCloud AI Store for Federated Learning in Biomedicine and Beyond
Machine Learning (ML) and Artificial Intelligence (AI) have shown promising results in many areas and are driven by the increasing amount of available data. However, this data is often distributed across different institutions and Machine Learning (ML) and Artificial Intelligence (AI) have shown promising results in many areas including biomedicine and are driven by the increasing amount of available data. However, patient-derived data is often distributed across different institutions and cannot be shared and pooled due to privacy concerns. Privacy-preserving methods, such as Federated Learning (FL), allow for training ML models on distributed data without sharing sensitive information. Despite this advantage, to date, the FL approach is only beginning to be used in biomedicine because the implementation of FL-based tools is time-consuming, and requires advanced programming skills. To remove this roadblock from the way of FL to biomedicine, we present FeatureCloud (https://featurecloud.ai), an all-in-one platform for the development and application of FL-based tools. FeatureCloud provides the developers with well-documented templates and a testing environment for the creation of FL-based apps. The apps which passed validation become available for the end-users in the FeatureCloud AI Store. The FeatureCloud apps are ready-to-use and require no programming skills, enabling FL for a broad community of biomedical researchers. To illustrate the utility of FeatureCloud, we present three FL-based tools: PARTEA (https://exbio.wzw.tum.de/partea/) for privacy-aware time-to-event analysis, Flimma (https://exbio.wzw.tum.de/flimma/) for federated differential expression analysis, and sPLINK (https://exbio.wzw.tum.de/splink) for secure genome-wide association studies (GWAS). We show that these federated apps produce similar results to their centralized versions applied to pooled data regardless of data distribution across the participants. It makes them a promising alternative to meta-analyses, whose results may be strongly affected by imbalanced target class or covariate distributions across the cohorts.
Joint Presentation with Dr. Olga Zolatareva
Mohammad Bakhtiari is a research assistant in the institute for computational bioinformatics at the University of Hamburg with a specialization in Federated Learning, Transfer Learning, Deep Learning, and SC-RNA Sequencing. He holds a Master’s degree in Artificial Intelligence and Robotics from the University of Teheran, Iran.