Interpreting convolutional neural networks in population genetics.
7 affiliations
- 1Institute for Bioinformatics and Medical Informatics (IBMI), University of Tübingen, 72076 Tübingen, Germany.
- 2Department of Biology, University of Pennsylvania, Philadelphia, PA, USA.
- 3Department of Computer Science, Haverford College, Haverford, PA, USA.
- 4Department of Genetics, University of North Carolina at Chapel Hill, USA.
- 5Cluster of Excellence "Controlling Microbes to Fight Infections", Mathematical and Computational Population Genetics, University of Tübingen, 72076 Tübingen, Germany.
- 6Cluster of Excellence "Machine Learning: New Perspectives for Science", University of Tübingen, Tübingen, Germany.
- 7Big Data Analytics in Bioinformatics, Justus-Liebig University, Giessen, Germany.
Abstract
Machine learning approaches have become a powerful alternative to traditional methods in population genetics. Convolutional neural networks (CNNs) in particular have been successful in inferring natural selection, recombination rates, introgression, dispersal distances, and effective population size changes. One limitation of CNNs and other deep learning methods is that they can be difficult to interpret. When they have been shown to be as or more successful than summary-statistic-based methods, what are they learning? Here, we investigate CNNs from two different methods: the pg-gan discriminator for identifying real vs. simulated data, and two networks trained to detect selective sweeps. We first compute correlations between learned network features and traditional summary statistics, then assess whether summary statistics can be predicted from the learned features. To understand the learned features, we compute feature importance through SHAP values and feature groupings through dimensionality reduction. Finally, we use decision trees and random forests to build an interpretable "model-of-the-model". Our results reveal that some CNN architectures can implicitly compute summary statistics such as pairwise heterozygosity, while statistics such as the site frequency spectrum are less similar to the network's learned features. We find that long-range linkage disequilibrium is readily approximated by the networks and may be more efficiently computed by CNNs than traditional methods (which are quadratic in the number of sites). Overall, this work contributes to the interpretability of deep learning methods in population genetics by clarifying the relationships between model architecture, learned network features, established summary statistics, and predicted evolutionary parameters.
Related
- Transfer Learning-Driven Convolutional Neural Networks for Classification of Plant Species from Leaf Vein Patterns
- Interpreting Graph Neural Networks for Antibacterial Drug Repurposing against Acinetobacter baumannii
- Interpreting graph neural networks reveals microbial association patterns in inflammatory bowel diseases
- Interpreting chemical crosslinks: Score-based approaches and deep neural networks
- Application of long short-term memory and convolutional neural networks for real-time bridge pier scour prediction
BibTeX
@article{xu2026interpreting,
title = {{Interpreting convolutional neural networks in population genetics.}},
author = {Huiting Xu and Leon Zong and Dylan D Ray and Lei Lei and Daniel R Schrider and Franz Baumdicker and Sara Mathieson},
journal = {Molecular biology and evolution},
year = {2026},
month = oct,
doi = {10.1093/molbev/msag256},
url = {https://europepmc.org/article/MED/42831439},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - Interpreting convolutional neural networks in population genetics.
AU - Huiting Xu
AU - Leon Zong
AU - Dylan D Ray
AU - Lei Lei
AU - Daniel R Schrider
AU - Franz Baumdicker
AU - Sara Mathieson
PY - 2026
DA - 2026/10/01
JO - Molecular biology and evolution
DO - 10.1093/molbev/msag256
UR - https://europepmc.org/article/MED/42831439
AB - Machine learning approaches have become a powerful alternative to traditional methods in population genetics. Convolutional neural networks (CNNs) in particular have been successful in inferring natural selection, recombination rates, introgression, dispersal distances, and effective population size changes. One limitation of CNNs and other deep learning methods is that they can be difficult to interpret. When they have been shown to be as or more successful than summary-statistic-based methods, what are they learning? Here, we investigate CNNs from two different methods: the pg-gan discriminator for identifying real vs. simulated data, and two networks trained to detect selective sweeps. We first compute correlations between learned network features and traditional summary statistics, then assess whether summary statistics can be predicted from the learned features. To understand the learned features, we compute feature importance through SHAP values and feature groupings through dimensionality reduction. Finally, we use decision trees and random forests to build an interpretable "model-of-the-model". Our results reveal that some CNN architectures can implicitly compute summary statistics such as pairwise heterozygosity, while statistics such as the site frequency spectrum are less similar to the network's learned features. We find that long-range linkage disequilibrium is readily approximated by the networks and may be more efficiently computed by CNNs than traditional methods (which are quadratic in the number of sites). Overall, this work contributes to the interpretability of deep learning methods in population genetics by clarifying the relationships between model architecture, learned network features, established summary statistics, and predicted evolutionary parameters.
SN - 0737-4038
SN - 1537-1719
ER -