The Perez Lab builds deep learning and Bayesian methods that turn genomes and phylogenies into histories: how lineages split, spread, adapt and go extinct, from Brazilian cacti and Amazonian fishes to malaria vectors and HIV.
Evolutionary biology and machine learning, inside one of Europe's oldest botanical gardens.
The lab is based at the Real Jardín Botánico in Madrid, a living museum with more than 265 years of history and a research centre of the Spanish National Research Council (CSIC). The garden's mission is to understand and conserve the biodiversity of plants, fungi and other organisms, and to train people to do the same at every level of education. We add a computational layer to that mission.
Our work sits on complex diversification processes: the recent, messy, incompletely sorted parts of the tree of life where species boundaries blur and demography leaves its clearest marks. To get at them we develop Bayesian and machine-learning methods, in particular neural networks trained on simulated data, for species delimitation, phylogeography, population genetics, and phylogenetic and diversification analyses.
Being inside a botanical garden shapes the questions. Pleistocene refugia in Neotropical cacti, admixture in the Canarian flora and the conservation of threatened endemics sit next to projects on Amazonian fishes, primate diversification and the transmission dynamics of pathogens such as HIV and the malaria vectors of the Anopheles gambiae complex. The same simulators and networks serve all of them.
Four threads, one toolbox. Each line pairs a biological question with the simulation-based inference needed to answer it.
Tree representations and neural architectures that let a network read a phylogeny directly, for model selection and parameter estimation in phylodynamics (BD, BDEI, BDSS) and macroevolutionary diversification (BiSSE and relatives). PhyloCNN is the current flagship.
Where coalescent methods and taxonomy disagree, we simulate under explicit demographic models and let CNNs or ABC decide: interglacial microrefugia in cactus complexes, island admixture in Macaronesia, and species complexes in the gray zone of speciation.
Combining population genomics with comparative cytogenetics in Neotropical fishes: karyotype turnover, neo-sex chromosomes and chromosomal rearrangements as drivers of divergence in arowanas, arapaima, trahiras and armored catfish.
Species distribution modelling and remote-sensing data to assess climate-change impacts on endangered species, and continental-scale reconstructions of responses to Quaternary change. The newest branch applies generative models to evolutionary inference in malaria vectors, and asks how AI should be used in science at all.
Everything is open. Scripts and notebooks accompany each paper so results can be re-run, re-trained and re-purposed.
Simulation, tree encoding, CNN training and posterior analysis for phylodynamics and diversification. Perez & Gascuel, Systematic Biology 2025.
↗ DeepIDIntegrative species delimitation with two coupled CNNs that read thousands of loci together with phenotypic traits.
↗ CNN_spDelimitationCoalescent-based species delimitation with deep learning in a fragmented cactus system. Mol. Ecol. Resour. 2022.
↗ CNN_DemographyArapaimaData simulation, demographic model selection and parameter estimation with CNNs. Ecography 2020.
↗ CNN_CanariasPhylogeographic model testing for the Canarian flora. Molecular Ecology 2024.
↗ Physalia_MLPopGenCourse materials: machine learning for population genetics (2023, 2025, 2026 editions).
↗# a typical run, start to finish $ python generate_parameters.py --model BDSS \ --R0 1.5,5 --tree-size 200,500 $ simulate_trees params.csv → trees.nwk $ python encode_trees.py trees.nwk → encoded.csv $ jupyter nbconvert --execute train_CNN.ipynb epoch 48/50 val_mae 0.041 ✓ $ python predict.py empirical_HIV.nwk R0 = 2.31 [1.94, 2.70] sampling p = 0.26 $
Peer-reviewed articles, preprints and book chapters. The full record lives on Google Scholar and ORCID.
Courses and workshops on simulation-based inference, from ABC in 2014 to deep learning today.
Recurring international course with Physalia Courses. Simulation, summary statistics, CNNs on genotype matrices and trees, and how to trust (or not) a trained network.
One-week course for the Genetics and Evolution graduate program: species distribution modelling, isolation by distance and by environment, population structure, all in R.
Lecture and practical at the NSF–SSB Phylogeography and Phylogenetics Workshop: simulating genetic data under demographic models and testing them with Approximate Bayesian Computation.
Lectures at the Phylogeography Workshop, Universidade Estadual de Campinas.
The lab is in the Real Jardín Botánico, next to the Prado, in central Madrid.
We welcome students and postdocs who want to work between evolutionary biology and machine learning, whatever side they are coming from. If you have a question about speciation, phylogeography or diversification and suspect a simulator plus a neural network could answer it, write to us with a short note on what you would like to do.
Real Jardín Botánico, CSIC
Plaza de Murillo 2, 28014 Madrid
40.4113° N · 3.6911° W
Metro Atocha or Banco de España; the public entrance is the Puerta de Murillo, opposite the Prado.
Open in Google Maps ↗The lab is young and growing. Students and collaborators will appear here as they join.
Manolo is a researcher at the Real Jardín Botánico in Madrid (CSIC). He trained as an evolutionary geneticist, earning his PhD at the Federal University of São Carlos in Brazil with work on the phylogeography of cacti from dry Neotropical environments, and has since moved steadily toward the computational side of the field.
His research develops Bayesian and machine-learning methods for population genetics, species delimitation, biogeography, and phylogenetic and diversification analyses, applied to datasets that range from macroevolution to epidemiology. He contributes to the stdpopsim consortium and teaches machine learning for population genetics internationally.
Before Madrid he held positions at Imperial College London (I-X, AI in science), the Muséum national d'Histoire naturelle in Paris, and the Fish Cytogenetics Lab at UFSCar.
Collection localities, in order. Each stop added a group of organisms, or a method, that the lab still works with.
The people the lab works with. Positions in the lab are listed under Contact.
Evolutionary biologist developing machine-learning methods to infer evolutionary dynamics from phylogenetic trees, with a particular focus on diversification and applications to outbreak dynamics. As a biologist developing methodological tools, he designs and evaluates these approaches around real biological questions and empirical datasets, while also applying established likelihood-based methods for phylogenetic reconstruction, molecular dating, diversification analysis and historical biogeography.