Son Hoang Nguyen

Son Nguyen, Adam Wang, Albert Montillo

Traditional deep learning (DL) models have two ubiquitous limitations. First, they assume training samples are independent and identically distributed (i.i.d), an assumption often violated in real-world datasets where samples have additional correlation due to repeat measurements (e.g., on the same participants in a longitudinal study or cells from the same sequencer). This leads to performance degradation, limited generalization, and covariate confounding, which induces Type I and Type II errors. Second, DL models typically prioritize overall accuracy, favoring accuracy on the majority while sacrificing performance for underrepresented subpopulations, leading to unfair, biased models. This is critical to remediate, particularly in models which influence decisions regarding loan approvals and healthcare. To address these issues, we propose the Fair Mixed Effects Deep Learning (Fair MEDL) framework. This framework quantifies cluster-invariant fixed effects (FE) and cluster-specific random effects (RE) through: 1) a cluster adversary for learning invariant FE, 2) a Bayesian neural network for RE, and 3) a mixing function combining FE and RE for final predictions. Fairness is enhanced through architectural and loss function changes introduced by an adversarial debiasing network. We formally define and demonstrate improved fairness across three metrics: equalized odds, demographic parity, and counterfactual fairness, for both classification and regression tasks. Our method also identifies and de-weights confounded covariates, mitigating Type I and II errors. The framework is comprehensively evaluated across three datasets spanning two industries, including finance and healthcare. The Fair MEDL framework improves fairness by 86.4% for Age, 64.9% for Race, 57.8% for Sex, and 36.2% for Marital status, while maintaining robust predictive performance.

Aixa X. Andrade, Son Nguyen, Austin Marckx et al.

Single-cell RNA sequencing enables high-resolution analysis of cellular heterogeneity, yet disentangling biological signal from batch effects remains a major challenge. Existing batch-correction algorithms suppress or discard batch-related variation rather than modeling it. We propose scMEDAL, single-cell Mixed Effects Deep Autoencoder Learning, a framework that separately models batch-invariant and batch-specific effects using two complementary subnetworks. The principal innovation, scMEDAL-RE, is a random-effects Bayesian autoencoder that learns batch-specific representations while preserving biologically meaningful information confounded with batch effects signal often lost under standard correction. Complementing it, the fixed-effects subnetwork, scMEDAL-FE, trained via adversarial learning provides a default batch-correction component. Evaluations across diverse conditions (autism, leukemia, cardiovascular), cell types, and technical and biological effects show that scMEDAL-RE produces interpretable, batch-specific embeddings that complement both scMEDAL-FE and established correction methods (scVI, Scanorama, Harmony, SAUCIE), yielding more accurate prediction of disease status, donor group, and tissue. scMEDAL also provides generative visualizations, including counterfactual reconstructions of a cell's expression as if acquired in another batch. The framework allows substitution of the fixed-effects component with other correction methods, while retaining scMEDAL-RE's enhanced predictive power and visualization. Overall, scMEDAL is a versatile, interpretable framework that complements existing correction, providing enhanced insight into cellular heterogeneity and data acquisition.