Research

We study evolutionary biology, ecology, cultural evolution, behavior, and communication, using mathematical, computational, statistical, and AI models and collaborations with empirical biologists. Our interests span various biological systems, from viruses and microbes to cetaceans and hominids.

Our long-term goal is to understand how mechanisms that generate and transmit variation at different levels-genetic, behavioral, cultural-interact to shape the evolutionary process.

To get involved in our research, please email Yoav.

Current research interests

  1. Evolutionary Genetics and Genomics. We investigate adaptation mechanisms, including stress-induced mutagenesis, aneuploidy, and other genetic processes driving evolutionary change.
  2. Cultural Evolution. We explore the dynamics of cultural transmission, social learning, and the evolution of cooperation in humans and other animals.
  3. Computational and Statistical Modeling. We develop statistical and deep learning models to analyze (i) evolutionary and ecological experiments and (ii) anthropological and archeological data.
  4. Microbial Evolution and Ecology. We study microbial growth, competition, and adaptation in mixed cultures, focusing on ecological and evolutionary processes.
  5. Language and Communication. We examine the emergence and evolution of language structures and communication systems, including those of humans, other animals, and artificial intelligence.

Support

Our work is supported by grants from the Israel Science Foundation (ISF), US–Israel Binational Science Foundation (BSF), Templeton Foundation, Minerva Stiftung, Tel Aviv University Data Science Center, Safra Center for Bioinformatics, Amazon, and NVIDIA.

Our research

Evolutionary genetics

We have been deeply interested in the evolution of modifier genes, which affect processes such as mutation, recombination, dispersal, and learning without directly affecting fitness (Ram 2024). We developed a theoretical basis to explain the evolution of stress-induced mutagenesis – the phenomena in which stress induces a transient increase in mutation rates. Stress-induced mutagenesis is prevalent in bacteria and many eukaryote species, from yeast to human cancer cells. We used mathematical models and computer simulations to show that (i) stress-induced mutagenesis is favored by natural selection (Ram & Hadany 2012); that (ii) this is also true in the presence of rare recombination (Ram & Hadany 2019); that (iii) stress-induced mutagenesis increases the rate of complex adaptation without reducing the mean fitness of the population (Ram & Hadany 2014); and that (iv) errors in regulation of mutagenesis can be compensated by population communication (Dellus-Gur et al. 2017).

We have also found a general principle underlying this phenomenon, which we have termed the Modified Mean Fitness Principle (Ram et al. 2018). It determines that if below-average individuals generate more variation (be it genetic or phenotypic) than above-average individuals, then the population mean fitness will increase. The mechanisms for the generation of variation are not limited to mutation but rather may include dispersal (Gueijman et al. 2013), social learning, etc. We have also investigated how phenotypic switching of the mutation rate affects evolution (Lobinska et al. 2023), and how strong recurrent mutation affects the probability and time of fixation of beneficial alleles (Pontz et al. 2023).

In a project supported by NVIDIA and Amazon, we developed self-replicating artificial neural networks (SeRANNs), capable of copying their own genotypes and performing classification tasks that determine their reproductive success. We evolved 1,000 SeRANNs over thousands of generations and observed evolutionary phenomena such as adaptation, clonal interference, epistasis, and the evolution of mutation rates. Our findings demonstrate that universal evolutionary dynamics can naturally emerge in self-replicating models with implicit selection and mutation processes (Shvartzman and Ram 2024).

In a project funded by the ISF, we have studied the phenomnon of aneuploidy and copy number variation. We developed an evolutionary model to examine the role of aneuploidy in yeast populations under heat stress. Our findings suggest that while aneuploidy can provide immediate fitness benefits in stressful environments, it may act as an evolutionary diversion rather than a stepping stone, potentially delaying adaptation by hindering the fixation of more refined beneficial mutations (Kohanovski et al. 2024). Next, we developed a mathematical model to assess how aneuploidy influences cancer cell populations under chemotherapy. Our findings indicate that aneuploidy can enhance tumor survival by increasing the likelihood of evolutionary rescue, particularly in smaller secondary tumors. This research underscores the significant role of chromosomal instability in cancer relapse following treatment (Stana et al. 2025).

Cultural evolution

We studied the evolution of oblique transmission, in which offspring inherit their phenotype from a non-parental adult rather than their parents. This occurs primarily during social learning, but also in symbiont and pathogen transmission. We found that the evolutionarily stable rate of oblique transmission differs markedly from the rate that maximizes the geometric mean fitness of the population, implying a form ofprisoner's dilemma for social learning (Ram, Liberman & Feldman 2018). We've also studied how fluctuations in the rate of vertical vs. oblique transmission affect trait polymorphism (Ram, Liberman & Feldman 2019), what happens when the rate of vertical vs. oblique transmission depends on trait frequency (Liberman et al. 2019), or when transmission is biased towards to the more (or less) common phenotype (Denton et al. 2020), how such transmission biases affect cumulative culture (Denton et al. 2022), and how cultural evolution proceeds under prestige bias (Egozi and Ram 2024).

We also examined the evolution of cooperative behavior under non-vertical cultural transmission, focusing on the effect of assortment between cultural transmission and social interactions (Cohen et al. 2021), the effect of conformity and non-conformity (Denton, Ram & Feldman 2021), the emergence of cooperative hunting (Borofsky, Feldman & Ram 2024)

Recently, we analyzed datasets to identify four classes of cultural features and distinct clusters within Austronesian societies. We found varying modes of transmission and patterns of variation across these cultural classes, with geography alone insufficient to explain the observed diversity. Our results demonstrate how methods inspired by population genetics can effectively analyze cultural transmission and variation (Macdonald et al. 2024a). To support such analyses, we also developed a Bayesian method for estimating the number of significant principal components in cultural datasets (Macdonald et al. 2024b).

We are also revisiting classical models of the transmission of human handedness. We reanalyzed McManus' genetic model of handedness using modern Bayesian inference (Oron et al. 2025), and reanalyzed the data underlying the gene-culture coevolutionary account of handedness transmission (Karstadt et al. 2026).

With Ran Barkai and colleagues, we consider how the collection and curation of souvenirs by early hominins may have created a cultural niche for the evolution of autobiographical memory and the sense of self (Barkai et al. 2026).

Behavioral Ecology

We developed and analyzed models for the evolution of menopause, which led us to propose the Mama's Boy hypothesis, according to which prolonged maternal investment in sons and grandsons (rather than in all grandchildren) can select for post-reproductive lifespan (Yosef et al. 2026).

Analysis of evolutionary experiments

We developed and tested a new method for predicting growth in a mixed culture solely from growth curve data (Ram et al. 2019a, Ram et al. 2019b) in collaboration with the Berman lab at Tel Aviv University, Cooper lab at University of Houston, and with Marc Feldman and Lilach Hadany. We validated this method by performing growth curve and competition experiments with bacteria. Our new method not only results in a simple and cost-effective approach for estimating growth in a mixed culture and inferring competitive fitness but also provides information on the specific growth traits that contribute to differences in fitness, thus helping to bridge the gap between ecology and evolution.

We also collaborated with experimental microbiologists to analyze the results of evolutionary experiments. In experiments performed by the Kupiec lab at Tel Aviv University, haploid yeast cells frequently became diploid via two distinct genetic mechanisms. Using approximate Bayesian approximation (ABC), we estimated the rate of endoreduplication in these experiments and found that it was much higher than the mutation rate (Harari et al. 2018a, Harari et al. 2018b). In a project funded by the BSF with the Gresham lab at New York University, we inferred the formation rate and fitness effect of beneficial copy number variation mutations at a specific locus in yeast growing in a chemostat. This was performed using a state-of-the-art neural-network-based method, NPE, and evolutionary simulations (Avecilla et al. 2022; Chuong et al. 2025; De et al. 2026). Because such experiments often produce highly variable replicates, we developed a collective posterior approach that combines simulation-based inference across replicates while accounting for the variation between them (Ben Nun et al. 2026). We also applied this novel technique to estimate the mutation rate of the MS2 bacteriophage from evolutionary experiments together with the Stern lab at Tel Aviv University (Capsi et al. 2023), and, with the same lab, to infer which viral proteins act as public goods during coinfection (Maoz et al. 2026).

Language and communication

With Limor Raviv and Lukas Galke, we study how the structure of a language affects the way it is learned, by humans and by machines. We found that deep neural networks, like humans, learn compositional languages faster and generalize from them better, suggesting that domain-general learning pressures can contribute to the emergence of linguistic structure (Galke et al. 2024). We have also outlined what is missing from emergent communication research if it is to inform theories of human language evolution (Galke et al. 2022).

With Inbal Arnon, Amiyaal Ilany, Mirjam Knörnschild, and Yossi Yovel, we examined a key challenge in using AI to decode non-human animal communication: models that identify acoustic patterns in vocalizations do not necessarily recover their meaning, since meaning depends on how a call is interpreted and acted upon by its receiver. Using pre-verbal human toddlers' vocalizations as a test case, where the interpretation of listeners can be assessed, we show that acoustic pattern recognition alone is not sufficient for decoding communicative meaning (Taub et al. 2026).

We are also using self-supervised learning to study rock hyrax vocalizations. Using animal2vec, a self-supervised model for animal sounds, we asked whether representations learned from other species' calls (meerkats, Egyptian fruit bats) can transfer to rock hyrax vocalizations. We find that self-supervised pretraining directly on hyrax recordings substantially improves individual identification compared with cross-species transfer, that meerkat-derived representations offer only a partial, non-additive advantage, and that transfer from fruit bat recordings is considerably less effective — indicating that learned representations do not generalize universally across species, tasks, and recording conditions (Zlotorinsky et al., in preparation).

Deep learning for ecology

In collaboration with the Geffen lab at Tel Aviv University, Uri Roll, and Sarig Gafny, we developed a photo-identification pipeline for the critically endangered Hula painted frog (Latonia nigriventer). Using zero-shot deep local-feature matching, without any species-specific training, we achieved near-perfect re-identification of individuals from field photographs, offering a non-invasive alternative to physical tagging for monitoring rare populations (Yesharim et al. 2026). We are now applying the same approach to photo-ID of the fire salamander (Yesharim et al., in preparation). We extended this work to the AnimalCLEF 2026 challenge, comparing global embeddings with local feature matching for animal re-identification across species and characterizing adaptive validation overfitting (Yesharim & Ram 2026).

COVID-19 pandemic

We used an SEIR model and MCMC parameter estimation to infer the effective start of non-pharmaceutical interventions (e.g., lockdowns) during the initial outbreaks of COVID-19 in 12 countries. We found that in the majority of cases, the effective start date is significantly different from the official and that this late effective start may lead to incorrect interpretation of the impact of interventions (Kohanovski et al. 2022).

We studied a model in which human behavior changes in response to the number of infected individuals causing the transmission rate to decrease. We found cycles, chaos, and high sensitivity to model parameters. This suggests that the predictive power of epidemiological models is hampered by complex human behavior (Arthur et al 2021; Stanford News).

We used evo-epidemiological models to examine the effect of non-pharmaceutical interventions (social distancing, lockdowns, etc.) and PCR testing on the evolution of virulence, attenuation, and test-evasiveness in SARS-CoV-2. We find conditions under which stronger measures select for less virulent strains, and more testing selects for less detectable strains, even if those strains are also less infectious (Gurevich et al. 2022).

Collaborators

Tel Aviv University

  • Lilach Hadany
  • Adi Stern
  • Judy Berman
  • Uri Liberman
  • Uri Obolski
  • Eli Geffen
  • Yoni Belmaker
  • Yossi Yovel
  • Martin Kupiec
  • Inon Scharf
  • Uri Ben-David
  • Amiyaal Ilany
  • Shay Rotics
  • Orr Spiegel
  • Ran Barkai

Elsewhere

  • Marc Feldman, Stanford University
  • David Gresham, New York University
  • Tzachi Pilpel, Weizmann Institute of Science
  • Oren Kolodny, Hebrew University of Jerusalem
  • Shay Covo, Hebrew University of Jerusalem
  • Shimon Harrus, Hebrew University of Jerusalem
  • Tim Cooper, Massey University
  • Daniel Weissman, Emory University
  • Anna Selmecki, University of Minnesota
  • Limor Raviv, Max Planck Institute for Psycholinguistics
  • Lee Altenberg, University of Hawaiʻi at Mānoa
  • Lukas Galke, University of Southern Denmark
  • Liran Samuni, German Primate Centre
  • Uri Roll, Ben-Gurion University of the Negev
  • Sarig Gafny, Ruppin Academic Center
  • Inbal Arnon, Hebrew University of Jerusalem