Search

Federated learning (FL) allows the collaborative training of AI models without needing to share raw data. This capability makes it especially interesting for healthcare applications where patient and data privacy is of utmost concern. However, recent works on the inversion of deep neural networks from model gradients raised concerns about the security of FL in preventing the leakage of training data. In this work, we show that these attacks presented in the literature are impractical in FL use-cases where the clients’ training involves updating the Batch Normalization (BN) statistics and provide a new baseline attack that works for such scenarios. Furthermore, we present new ways to measure and visualize potential data leakage in FL. Our work is a step towards establishing reproducible methods of measuring data leakage in FL and could help determine the optimal tradeoffs between privacy-preserving techniques, such as differential privacy, and model accuracy based on quantifiable metrics.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site’s data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

ABSTRACTIgG4, a common type of therapeutic antibody, is less stable during manufacturing processes compared with IgG1. Aggregation and fragmentation are the two main challenges. Here, we report instability of the heavy chain (HC) C-terminal region under acidic conditions, which leads to cleavage and aggregation. Leu445, at the C-terminal region of the HC in IgG4,plays a critical role in its acid-induced fragmentation and subsequent aggregation. We found that mutating HC C-terminal Leu445to Pro (the corresponding residue in IgG1) in IgG4_CDR-X significantly reduces fragmentation and aggregation, while mutating Pro445to Leu in IgG1_CDR-X promotes fragmentation and aggregation. HC C-terminal Gly446cleavage was observed in low pH citrate buffer and resulted in further fragmentation and aggregation, whereas, glycine buffer can completely inhibit the cleavage and aggregation. It is proposed that cleavages occur through acid-induced hydrolysis under acidic conditions and glycine stabilizes IgG4via two main mechanisms: 1) product feedback inhibition of the hydrolysis reaction, and 2) stabilization of protein conformation by direct interaction with the peptide backbone and charged side chains. Experiments using IgG4molecules IgG4_CDR-Y and IgG4_CDR-Z with the same CH domains as IgG4_CDR-X, but different complementarity-determining regions (CDRs), indicate that the stability of the HC C-terminal region is also closely related to the sequence of the CDRs. The stability of IgG4_CDR-X is significantly improved when binding to its target. Both observations suggest that there are potential interactions between Fab and CH2-CH3 domains, which could be the key factor affecting the stability of IgG antibodies.