SMDNet deep learning framework guides design of Aβ42-targeted fluorescent probes for Alzheimer's disease

An AI framework combining self-training and multimodal data designed Aβ42-targeted fluorescent probes. The model identified a candidate (TA3) with favorable binding affinity and high-contrast imaging performance, validated across multiple scaffold classes (exact performance metr…

• SMDNet uses iterative self-training with confidence- and distribution-aware sampling, integrating molecular graphs, fingerprints, and protein descriptors via cross-attention and adaptive layer normalization.
• Ablation studies, external validation, and generalization analyses confirmed strong predictive ability; interpretability highlighted chemically meaningful substructures.
• Proof-of-concept restricted to in silico design and in vitro validation of five ThT-derived candidates; clinical translation and in vivo performance remain unassessed.

RadPigeon summaries are original and for information only. They are not clinical advice.