Genome modeling and design across all domains of life with Evo 2

bioRxiv (Cold Spring Harbor Laboratory) · 2025-02-21 · 190 citations

Abstract All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedented 1 million token context window with single-nucleotide resolution. Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions. Beyond its predictive capabilities, Evo 2 generates mitochondrial, prokaryotic, and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Guiding Evo 2 via inference-time search enables controllable generation of epigenomic structure, for which we demonstrate the first inference-time scaling results in biology. We make Evo 2 fully open, including model parameters, training code, inference code, and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.

Effector CD8 T cell differentiation in primary and breakthrough SARS-CoV-2 infection in mice

UNC Libraries · 2025-03-20

The nature of the effector and memory T cell response in the lungs following acute SARS-CoV-2 infections remains largely unknown. To define the pulmonary T-cell response to COVID-19, we compared effector and memory T-cell responses to SARS-CoV-2 and influenza A virus (IAV) in mice. Both viruses elicited potent effector T cell responses in lungs, but memory T cells showed exaggerated contraction in SARS-CoV-2-infected mice. Specifically, unlike the T-bet/EOMES-driven effector transcription program in IAV lungs, SARS-CoV-2-specific CD8 T cells embarked on a STAT-3-centric transcriptional program, a defining characteristic of a pro-fibro-inflammatory program: limited cytotoxicity, diminished expression of tissue-protective inhibitory receptors (PD-1, LAG-3, and TIGIT), and augmented mucosal imprinting (CD103). Circulating CD45RO+HLA-DR+ CD8 T cells in hospitalized COVID-19 patients expressed elevated levels of STAT-3 and low levels of TIGIT. IL-6 blockade experiments implicated IL-6 in STAT-3 induction and downregulation of PD-1 expression on SARS-CoV-2-specific primary effector CD8 T cells. Memory CD8 T cells specific to a single epitope, induced by mucosal vaccination, differentiated into cytotoxic effectors and expressed high levels of CD103, effectively reducing viral burden in lungs following a breakthrough SARS-CoV-2 infection. Our findings have implications for developing targeted immunotherapies to mitigate immunopathology and promote protective T cell immunity to SARS-CoV-2.

Predicting cellular responses to perturbation across diverse contexts with State

bioRxiv (Cold Spring Harbor Laboratory) · 2025-06-27 · 53 citations

Abstract Cellular responses to perturbations are a cornerstone for understanding biological mechanisms and selecting drug targets. While machine learning models offer tremendous potential for predicting perturbation effects, they currently struggle to generalize to unobserved cellular contexts. Here, we introduce S tate , a transformer model that predicts perturbation effects while accounting for cellular heterogeneity within and across experiments. S tate predicts perturbation effects across sets of cells and is trained using gene expression data from over 100 million perturbed cells. S tate improved discrimination of effects on large datasets by more than 30% and identified differentially expressed genes across genetic, signaling and chemical perturbations with significantly improved accuracy. Using its cell embedding trained on observational data from 167 million cells, S tate identified strong perturbations in novel cellular contexts where no perturbations were observed during training. We further introduce Cell-Eval, a comprehensive evaluation framework that highlights S tate ’s ability to detect cell type-specific perturbation responses, such as cell survival. Overall, the performance and flexibility of S tate sets the stage for scaling the development of virtual cell models.

Virtual Cell Challenge: Toward a Turing test for the virtual cell

Cell · 2025-06-01 · 49 citations

Effector CD8 T cell differentiation in primary and breakthrough SARS-CoV-2 infection in mice

Communications Biology · 2025-03-08 · 2 citations

The nature of the effector and memory T cell response in the lungs following acute SARS-CoV-2 infections remains largely unknown. To define the pulmonary T-cell response to COVID-19, we compared effector and memory T-cell responses to SARS-CoV-2 and influenza A virus (IAV) in mice. Both viruses elicited potent effector T cell responses in lungs, but memory T cells showed exaggerated contraction in SARS-CoV-2-infected mice. Specifically, unlike the T-bet/EOMES-driven effector transcription program in IAV lungs, SARS-CoV-2-specific CD8 T cells embarked on a STAT-3-centric transcriptional program, a defining characteristic of a pro-fibro-inflammatory program: limited cytotoxicity, diminished expression of tissue-protective inhibitory receptors (PD-1, LAG-3, and TIGIT), and augmented mucosal imprinting (CD103). Circulating CD45RO+HLA-DR+ CD8 T cells in hospitalized COVID-19 patients expressed elevated levels of STAT-3 and low levels of TIGIT. IL-6 blockade experiments implicated IL-6 in STAT-3 induction and downregulation of PD-1 expression on SARS-CoV-2-specific primary effector CD8 T cells. Memory CD8 T cells specific to a single epitope, induced by mucosal vaccination, differentiated into cytotoxic effectors and expressed high levels of CD103, effectively reducing viral burden in lungs following a breakthrough SARS-CoV-2 infection. Our findings have implications for developing targeted immunotherapies to mitigate immunopathology and promote protective T cell immunity to SARS-CoV-2. SARS-CoV-2 infection triggers a potent but atypical CD8 T-cell response in lungs; recall responses of CD8 + T cells specific to a single epitope (SIINFEKL) induced by vaccination can effectively reduce viral loads following breakthrough SARS-CoV-2 infection.

Sequence modeling and design from molecular to genome scale with Evo

bioRxiv (Cold Spring Harbor Laboratory) · 2024 · 81 citations

• Computer Science
• Computational biology
• Biology

The genome is a sequence that completely encodes the DNA, RNA, and proteins that orchestrate the function of a whole organism. Advances in machine learning combined with massive datasets of whole genomes could enable a biological foundation model that accelerates the mechanistic understanding and generative design of complex molecular interactions. We report Evo, a genomic foundation model that enables prediction and generation tasks from the molecular to genome scale. Using an architecture based on advances in deep signal processing, we scale Evo to 7 billion parameters with a context length of 131 kilobases (kb) at single-nucleotide, byte resolution. Trained on whole prokaryotic genomes, Evo can generalize across the three fundamental modalities of the central dogma of molecular biology to perform zero-shot function prediction that is competitive with, or outperforms, leading domain-specific language models. Evo also excels at multi-element generation tasks, which we demonstrate by generating synthetic CRISPR-Cas molecular complexes and entire transposable systems for the first time. Using information learned over whole genomes, Evo can also predict gene essentiality at nucleotide resolution and can generate coding-rich sequences up to 650 kb in length, orders of magnitude longer than previous methods. Advances in multi-modal and multi-scale learning with Evo provides a promising path toward improving our understanding and control of biology across multiple levels of complexity.

Sequence modeling and design from molecular to genome scale with Evo

Science · 2024-11-14 · 394 citations

The genome is a sequence that encodes the DNA, RNA, and proteins that orchestrate an organism's function. We present Evo, a long-context genomic foundation model with a frontier architecture trained on millions of prokaryotic and phage genomes, and report scaling laws on DNA to complement observations in language and vision. Evo generalizes across DNA, RNA, and proteins, enabling zero-shot function prediction competitive with domain-specific language models and the generation of functional CRISPR-Cas and transposon systems, representing the first examples of protein-RNA and protein-DNA codesign with a language model. Evo also learns how small mutations affect whole-organism fitness and generates megabase-scale sequences with plausible genomic architecture. These prediction and generation capabilities span molecular to genomic scales of complexity, advancing our understanding and control of biology.

(103) 3-Piece Inflatable Penile Implant Pump Malfunction: Management using Pump Replacement Alone

The Journal of Sexual Medicine · 2023-05-01 · 1 citations

Abstract Introduction Expert opinion is that when an implant malfunctions after 2 years it should be replaced in its entirety, despite the fact that reservoir removal is associated with serious potential complications. Furthermore, modern reservoir construction is such that material fracture and leak is rare. Objective In this time we reviewed our experience with replacing only the reservoir in patients with malfunctioning 3-piece inflatable implants resulting from pump breakdown. Methods The study population consisted of (i) men who had had 3-piece inflatable penile implant surgery (ii) with device malfunction (iii) resulting from pump malfunction (iv) who had all other components functional (v) who underwent pump replacement alone and (vi) had ≥5 years follow-up after device repair (clinic visit, telephone call, letter). Only patients who had undergone a single prior implant operation were included. At the time of the repair, the pump was removed and replaced with a new pump, making sure that both cylinders and reservoir were filled fully intraoperatively and retained all fluid that was instilled. At the postoperative appointment, all patients were retrained in the correct means of device inflation, specially focused on avoidance of pump torqueing. Results 88 patients had pump tubing blow-out upon delivery of the pump at a mean duration post-primary implant of 9 ± 11 (2-16) years. Mean age 62 ±14 years. All patients had either a Mentor Alpha-1 or a Coloplast Titan devices placed. Tubing disruption was present ≤1cm from pump itself in 66 (75%) or >1cm from the pump in 22 (20%) patients. 25% had a blow out of the pump-reservoir tubing, while 66 (75%) had a tear in the pump-cylinder tubing (46 frayed tubing, 20 compete transection), evenly distributed between right and left-sided tubing. With a mean follow-up of 7±2.5 years, none of the devices had failed again. Conclusions In patients with a malfunctioning penile implant with proven pump-only failure, replacing the pump is a viable repair option, thus, circumventing the need for reservoir placement. Disclosure No

Extracellular Vesicle–associated GARP/TGFβ:LAP Mediates “Infectious” Allo-tolerance

Transplantation Direct · 2023-05-24 · 3 citations

Background. Here we test the hypothesis that, like CD81-associated “latent” IL35, the transforming growth factor (TGF)β:latency-associated peptide (LAP)/glycoprotein A repetitions predominant (GARP) complex was also tethered to small extracellular vesicles (sEVs), aka exosomes, produced by lymphocytes from allo-tolerized mice. Once these sEVs are taken up by conventional T cells, we also test whether TGFβ could be activated suppressing the local immune response. Methods. C57BL/6 mice were tolerized by i.p. injection of CBA/J splenocytes followed by anti-CD40L/CD154 antibody treatment on days 0, 2, and 4. On day 35, spleen and lymph nodes were extracted and isolated lymphocytes were restimulated with sonicates of CBA splenocytes overnight. sEVs were extracted from culture supernatants by ultracentrifugation (100 000 g ) and assayed for (a) the presence of TGFβ:LAP associated with tetraspanins CD81,CD63, and CD9 by enzyme-linked immunosorbent assay; (b) GARP, critical to membrane association of TGFβ:LAP and to activation from its latent form, as well as various TGFβ receptors; and (c) TGFβ-dependent function in 1° and 2° immunosuppression of tetanus toxoid-immunized B6 splenocytes using trans-vivo delayed–type hypersensitivity assay. Results. After tolerization, CBA-restimulated lymphocytes secreted GARP/TGFβ:LAP-coated extracellular vesicles. Like IL35 subunits, but unlike IL10, which was absent from ultracentrifuge pellets, GARP/TGFβ:LAP was mainly associated with CD81 + exosomes. sEV-bound GARP/TGFβ:LAP became active in both 1° and 2° immunosuppression, the latter requiring sEV uptake by “bystander” T cells and reexpression on the cell surface. Conclusions. Like other immune-suppressive components of the Treg exosome, which are produced in a latent form, exosomal GARP/TGFβ:LAP produced by allo-specific regulatory T cells undergoes either immediate activation (1° suppression) or internalization by naive T cells, followed by surface reexpression and subsequent activation (2°), to become suppressive. Our results imply a membrane-associated form of TGFβ:LAP that, like exosomal IL35, can target “bystander” lymphocytes. This new finding implicates exosomal TGFβ:LAP along with Treg-derived GARP as part of the infectious tolerance network.

(202) Exploring Erectile Hemodynamics in Men Complaining of Hard-Flaccid Syndrome (HFS)

The Journal of Sexual Medicine · 2023-05-01

Abstract Introduction There is increasing awareness of a condition, vernacularly termed HFS, in which men complain of subjective and palpable abnormalities in the flaccid penis (hardness, heaviness) often associated with erectile dysfunction (ED). Objective We attempted to define erectile hemodynamic profiles in this population. Methods The study population consisted of (i) males ≥18 years of age (ii) who presented with self-diagnosed HFS (iii) who underwent comprehensive history and examination (including genital and lower extremity neurological) and (iv) penile duplex Doppler ultrasound (PDDU). PDDU was performed using redosing of vasodilators. PSV ≥35 cm/s and EDV ≤3 were considered normal. Patients with Peyronie’s disease or a history of intracavernosal injection therapy were excluded. Results 88 men were analyzed. Mean age was 28 ± 12 years. Median number of vascular comorbidities was zero (0, 1). 66 (75%) had a history of depression or anxiety, 60 of whom had used pharmacotherapy for such at some time and 40 were actively using anxiolytics or anti-depressants at time of evaluation. No patient had a history of neurological conditions and neurological exam was normal in all. Mean duration of symptoms was 14 ± 20 months. Median number of prior physician consultations was 1 (1,4). All complained of impaired erectile function, 50% being incapable of having penetrative sexual relations. 91% had or were using PDE5 inhibitors. On examination all patients had a hypertonic flaccid penis with excellent stretch on traction. At PDDU, median number of penile injections was 2 (1, 3). 50% of men required phenylephrine reversal. All patients had normal PSV with 2/88 having abnormal EDV values, both of whom required reversal. Mean PSV was 45 ± 35 cm/s, mean EDV was 2 ± 11 cm/s. No patient had elevated echogenicity on B-mode scanning. Conclusions In this analysis, all men presenting with self-reported HFS had normal hemodynamics and an absence of fibrosis, suggesting the absence of vascular or gross cavernosal smooth muscle structural changes. Disclosure No