About

Dmitri Petrov is the Michelle and Kevin Douglas Professor in the School of Humanities and Sciences at Stanford University, within the Department of Biology. His primary research interests include the evolution of genomes and population genomics of adaptation and variation. He is involved in various affiliated programs such as Bio-X, CEHG, ChEM-H, Hopkins Marine Station, Woods Institute, and Wu Tsai Neuro. His work focuses on understanding evolutionary processes at the genomic level, contributing to the fields of evolutionary biology and genomics.

Research topics

• Biology
• Evolutionary biology
• Genetics
• Computer Science
• Demography
• Computational biology
• Statistics
• Ecology
• Neuroscience
• Geography

Selected publications

• Table S9 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Table 9 shows that number of sgNT1 tumors sampled in Cas9-EGFP V1 cohort for the Lenti-sgTSG19 and Lenti-sgTSG75 pools, Kras control V1 cohorts, Cas9-EGFP V1 cohort in the drug experiment and Kras control V1 cohort in the drug experiment.</p>

  PublisherDOI

• Table S3 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Table 3 shows the information of the sgRNA vectors in the Lenti-sgTSG19-sgV1/Cre, Lenti-sgTSG19-sgV3/Cre, Lenti-sgTSG75-sgV1/Cre, Lenti-sgTSG75-sgV3/Cre pools. Each gene was targeted by two to four sgRNAs. sgRNA sequence, source and sgID are indicated.</p>

  PublisherDOI

• Figure S9 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Figure 9 shows a schematic of the experimental workflow to isolate Setd2-proficient and Setd2-deficient V1- and V3-driven cancer cells, quality control metrics of the sorted cells, principal components analyses, and a heatmaps of differentially expressed genes.</p>

  PublisherOA PDFDOI

• Table S2 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Table 2 shows adaptively sampled mean (ASM) and adaptively sampled 95th percentile size (AS95) and significance of genes in the Lenti-sgTSG19 pools on EML4-ALK-V1 and V3-driven lung tumorigenesis.</p>

  PublisherDOI

• Figure S11 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Figure 11 shows the schematic for ATAC-Seq analysis of Eml4-Alk V1 and V3 lung tumors with and without Setd2 (sgSetd2) inactivation in vivo, principal component analyses, a heatmap of region accessibility across all samples, and HOMER Motif analyses.</p>

  PublisherOA PDFDOI

• Integrating noninvasive genetics and SECR to estimate snow leopard population in Pakistan

  Biological Conservation · 2026-01-26 · 1 citations

  article
  PublisherDOI

• Genotype-fitness mapping of adaptive mutants reveals shifting low-dimensional structure across divergent environments

  PLoS Biology · 2026-03-26 · 1 citations

  articleOpen accessSenior author

  A central goal in evolutionary biology is to predict the effect of a genetic mutation on fitness. This is a major challenge because it requires knowledge of both the phenotypic effects of a mutation and their importance in an arbitrary environment, which are high-dimensional quantities and difficult to guess a priori. Here, we address this problem by taking a top-down, data-driven approach to infer the mapping between genotypes, latent phenotypes, and fitness. We measure the fitness effects of a large collection of adaptive yeast mutants in many lab environments, from which we build low-dimensional, linear fitness landscapes. We find that these models are highly predictive of fitness variation for thousands of adaptive mutants, both in environments similar to where they evolved and also in divergent environments. This implies that the underlying genotype-phenotype-fitness maps for these adaptive mutants tend to be broadly low-dimensional. We further demonstrate that these maps only partially overlap across divergent environments, suggesting that the phenotypic determinants of fitness shift with the environment but remain low-dimensional. These results combine to emphasize the importance of environmental context in evolution, and suggest that top-down, low-dimensional fitness landscapes pave the way for evolutionary prediction.

  PublisherDOI

• Learning the Shape of Evolutionary Landscapes: Geometric Deep Learning Reveals Hidden Structure in Phenotype-to-Fitness Maps

  PRX Life · 2026-01-16 · 1 citations

  articleOpen accessSenior author

  Elucidating the complex relationships between genotypes, phenotypes, and fitness remains one of the fundamental challenges in evolutionary biology. Part of the difficulty arises from the enormous number of possible genotypes and the lack of understanding of the underlying phenotypic differences driving adaptation. Here we present a computational method that takes advantage of modern high-throughput fitness measurements to learn a map from high-dimensional fitness profiles to a low-dimensional latent space in a geometry-informed manner. We demonstrate that our approach using a Riemannian Hamiltonian variational autoencoder outperforms traditional linear dimensionality reduction techniques by capturing the nonlinear structure of the phenotype-fitness map. When applied to simulated adaptive dynamics, we show that the learned latent space retains information about the underlying adaptive phenotypic space and accurately reconstructs complex fitness landscapes. We then apply this method to a dataset of high-throughput fitness measurements of under different antibiotic pressures and demonstrate superior predictive power for out-of-sample data compared to linear approaches. Our work provides a data-driven implementation of Fisher's geometric model of adaptation, transforming it from a theoretical framework into an empirically grounded approach for understanding evolutionary dynamics using modern deep learning methods.

  PublisherDOI

• Figure S5 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Figure 5 shows the schematic of Lenti-sgTSG75-sgV1/Cre and Lenti-sgTSG75-sgV3/Cre pools, tumor initiation in Cas9-negative mice to determine the representation of Lenti-sgRNA/Cre vectors, and additional data from broader screen of tumor suppressor function in V1- and V3-driven lung cancer including tumor burden, tumor number and effect.</p>

  PublisherDOI

• Table S8 from EML4−ALK Variant-Specific Genetic Interactions Shape Lung Tumorigenesis

  2026-01-12

  articleOpen access

  <p>Supplementary Table 8 shows the Odds ratios (OR) and P-values from two-sided Fisher's exact tests comparing the occurrence of alterations in samples with EML4-ALK V1 relative to samples with EML4-ALK V3 for genes with at least 10 alterations across both cohorts</p>

  PublisherDOI

Recent grants

• Genomics of rapid adaptation in the lab and in the wild

  NIH · $705k · 2016–2021

• NIH Grant R01CA207133

  NIH · $1.7M · 2020

• (PQ4) Quantitative and multiplexed analysis of gene function in cancer in vivo

  NIH · $2.3M · 2018–2024

• NIH Grant R01GM100366

  NIH · $3.3M · 2021

• NIH Grant R01GM089926

  NIH · $969k · 2014

Frequent coauthors

• Monte M. Winslow

  Stratford University

  81 shared

• Hongchen Cai

  Stanford University

  58 shared

• Laura Andrejka

  Stanford University

  54 shared

• Chuan Li

  Peking University People's Hospital

  51 shared

• Josefa González

  Institut de Biologia Evolutiva

  44 shared

• Anna-Sophie Fiston-Lavier

  Centre National de la Recherche Scientifique

  36 shared

• Alan O. Bergland

  University of Virginia

  36 shared

• Ellie E. Armstrong

  Washington State University

  35 shared