Deep learning for mass extinction detection on fossilized phylogenies: power, limitations and lessons for simulation-based birth–death inference
Abstract
Detecting mass extinction events from phylogenies is a fundamental yet challenging task. While traditional likelihood-based methods are available, deep learning offers a powerful, simulation-based alternative. Here, we evaluate a deep learning approach using a novel hybrid model that combines graph neural networks with long short-term memory networks. This model analyses phylogenies—containing both extant species and fossils—simulated under a complex skyline fossilized birth–death model that incorporates mass extinctions and fluctuating background rates. We validate the architecture’s effectiveness through ablation studies. Our investigation revealed that the stochasticity of the simulation was a primary obstacle, creating significant ‘label noise’ that initially limited performance. A direct comparison showed our deep learning approach performed slightly better than Bayesian methods. It is robust to uncertainty in phylogenetic branch lengths and topology and generalizes to larger trees, but its performance degrades under model mismatch with higher background extinction rates. However, our work highlights a critical limitation: the model is highly specific to the definition of mass extinction it was trained on. Consequently, any modification to this definition necessitates retraining a new model from scratch. We conclude by summarizing the challenges and lessons learned for simulation-based inference in phylogenetic birth–death models.
Related
- Estimating Divergence Times and Diversification Rates with the Unresolved Fossilized Birth-Death Process
- Phylogenies as graphs: structured neural networks improve host origin predictions from paramyxovirus sequences
- How Phylogenies Inferred from 46 Mitochondrial Genomes Informed the Systematics of the Australasian Genus of Ticks, Bothriocroton, with the Description of Three New Species.
- Described eukaryotic species in published species-level phylogenies: placement by molecular data, dates and date uncertainty
- Gaussian processes on phylogenies: a kernel framework for building, checking and comparing models of trait evolution
BibTeX
@article{du2026deep,
title = {{Deep learning for mass extinction detection on fossilized phylogenies: power, limitations and lessons for simulation-based birth–death inference}},
author = {Minghao Du and Wenhui Wang and Jingqiang Tan and Joëlle Barido-Sottani},
journal = {Royal Society Open Science},
year = {2026},
month = jun,
doi = {10.1098/rsos.252096},
url = {https://hal.science/hal-05780792v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Deep learning for mass extinction detection on fossilized phylogenies: power, limitations and lessons for simulation-based birth–death inference
AU - Minghao Du
AU - Wenhui Wang
AU - Jingqiang Tan
AU - Joëlle Barido-Sottani
PY - 2026
DA - 2026/06/10
JO - Royal Society Open Science
DO - 10.1098/rsos.252096
UR - https://hal.science/hal-05780792v1
AB - Detecting mass extinction events from phylogenies is a fundamental yet challenging task. While traditional likelihood-based methods are available, deep learning offers a powerful, simulation-based alternative. Here, we evaluate a deep learning approach using a novel hybrid model that combines graph neural networks with long short-term memory networks. This model analyses phylogenies—containing both extant species and fossils—simulated under a complex skyline fossilized birth–death model that incorporates mass extinctions and fluctuating background rates. We validate the architecture’s effectiveness through ablation studies. Our investigation revealed that the stochasticity of the simulation was a primary obstacle, creating significant ‘label noise’ that initially limited performance. A direct comparison showed our deep learning approach performed slightly better than Bayesian methods. It is robust to uncertainty in phylogenetic branch lengths and topology and generalizes to larger trees, but its performance degrades under model mismatch with higher background extinction rates. However, our work highlights a critical limitation: the model is highly specific to the definition of mass extinction it was trained on. Consequently, any modification to this definition necessitates retraining a new model from scratch. We conclude by summarizing the challenges and lessons learned for simulation-based inference in phylogenetic birth–death models.
SN - 2054-5703
SN - 2054-5703
ER -