SEANN: A Domain-Informed Neural Network for Epidemiological Insights

Domain knowledge improves AI reliability in healthcare: SEANN demonstrates that integrating scientific consensus from meta-analyses directly into neural networks produces more generalizable predictions and scientifically plausible results than standard deep learning approaches, suggesting that responsible AI in epidemiology should actively incorporate existing domain expertise rather than treating it as separate from the model.

AI can work with limited medical data: The paper addresses a critical real-world constraint in healthcare—the scarcity of high-quality data—by showing that neural networks enhanced with domain knowledge perform better in these challenging conditions, which is important for developing trustworthy AI systems for rare diseases or under-resourced research areas.

Explainability matters for scientific credibility: By combining neural networks with explainable AI (XAI) tools and domain-informed constraints, SEANN produces relationships that are not only predictive but also scientifically plausible, highlighting that responsible AI in medicine must prioritize interpretability so that clinicians and researchers can understand and validate the model's reasoning.

Quantifying scientific consensus improves AI alignment: The use of Pooled Effect Sizes as a formal mechanism to encode scientific consensus into the learning process demonstrates a practical method for aligning AI systems with established knowledge, reducing the risk of models making recommendations that contradict well-established medical evidence.