publications

Reincke SM, Espinosa C, Chung P, James T, Berson E, Aghaeepour N. Journal of the American Medical Informatics Association. 2025 May 32(5), 920–927.

In this paper we highlight a challenge in AI disease prediction using Electronic Health Records (EHRs) called outcome conflation, where single-outcome models can lack specificity and incorrectly predict one disease due to shared features with others. This can lead to significant false-positive predictions and inappropriate clinical interventions, increasing both economic and psychological burdens. To overcome this, we developed MultiRiskNet, a novel multi-class AI architecture that simultaneously predicts multiple diseases. Our research demonstrates that MultiRiskNet dramatically improves prediction specificity and differentiation between correlated outcomes, substantially reducing false-positive predictions for conditions like pancreatic and ovarian cancer, thus paving the way for safer and more effective disease screening strategies.

Reincke SM*, von Wardenburg N*, Homeyer MA*, Kornau HC, Spagni G, Li LY, Kreye J, Sánchez-Sendín E, Blumenau S, Stappert D, Radbruch H, Hauser AE, Künkele A, Edes I, Schmitz D, Prüss H.Cell. 2023 Nov 9;186(23):5084-5097.e18. (* equal contributions)

In this paper we develop NMDAR-specific Chimeric Autoantibody Receptor (NMDAR-CAAR) T cells, a novel, precision therapy designed to treat NMDAR encephalitis, the most common autoimmune encephalitis. This approach allows these engineered T cells to selectively target and eliminate only the B cells producing disease-causing autoantibodies, leaving protective B cells unharmed. In extensive preclinical studies, NMDAR-CAAR T cells effectively depleted anti-NMDAR B cells, reduced pathogenic autoantibody levels in serum and brain, and demonstrated no signs of off-target toxicity. This targeted therapy aims to overcome the side effects of current broad immunosuppressive treatments, which often lead to chronic immunosuppression and associated complications, potentially leading to safer and more effective outcomes for patients.

Reincke SM*, Yuan M*, Kornau HC*, Corman VM*, van Hoof S, Sánchez-Sendin E, Ramberger M, Yu W, Hua Y, Tien H, Schmidt ML, Schwarz T, Jeworowski LM, Brandl SE, Rasmussen HF, Homeyer MA, Stöffler L, Barner M, Kunkel D, Huo S, Horler J, von Wardenburg N, Kroidl I, Eser TM, Wieser A, Geldmacher C, Hoelscher M, Gänzer H, Weiss G, Schmitz D, Drosten C, Prüss H, Wilson IA, Kreye J. Science. 2022 Feb 18;375(6582):782-787. (* equal contributions)

In this paper we investigated the antibody response elicited by SARS-CoV-2 Beta variant infection, a Variant of Concern known for resisting neutralization by wild-type antibodies. Our study revealed that Beta-infected patients developed lineage-specific antibodies that primarily target the Beta variant, showing reduced cross-neutralization of the wild-type virus. We also identified cross-reactive antibodies which offer broader protection against other variants. These findings emphasize the dynamic nature of the immune response to pathogens and have important implications for designing next-generation vaccines.