DendroNet

DendroNet integrates phylogenetic trees with neural networks to model genotype–phenotype relationships and improve generalization on phylogenetically structured biological datasets.

Key Features:

• Phylogenetic Awareness: Incorporates evolutionary relationships among samples by leveraging phylogenetic trees to account for non‑iid structure in training and testing datasets.
• Model Evolution Along Phylogenetic Branches: Enables neural network parameters or models to evolve along phylogenetic tree branches to accommodate shifts in genotype–phenotype mappings across lineages.
• Improved Generalization: Mitigates out-of-distribution generalization problems on phylogenetically structured data, yielding more reliable predictions across related but distinct evolutionary contexts.

Scientific Applications:

• Antibiotic Resistance Prediction: Applied to bacteria to improve accuracy of antibiotic resistance prediction relative to non‑phylogenetic approaches.
• Trophic Level Prediction in Fungi: Predicts trophic levels in fungal species to infer ecological roles and interactions.

Methodology:

Leverages phylogenetic trees to parameterize neural networks that evolve along tree branches and trains/evaluates models on phylogenetically structured training and testing datasets; validated via simulations and applications to bacterial antibiotic resistance and fungal trophic-level prediction.