EvoR: Genomic Foundation Models in R

EvoR is a lightweight R interface for the Evo2 Genomic Foundation Model (developed by the Arc Institute) via the NVIDIA BioNeMo API. It allows researchers to perform sequence generation, variant scoring, and high-dimensional embedding extraction natively in R.

---

🚀 Key Features

• Generation (evo2_query): Generate biological sequences (up to 1M bp) using deterministic or stochastic sampling.
• Scoring (evo2_get_score): Compute nucleotide-level log-probabilities for Variant Effect Prediction (VEP).
• Embeddings (evo2_query_embeddings, evo2_get_embeddings): Extract 4096-dimensional representation vectors for downstream ML modeling.

---

📦 Installation

Install the development version directly from GitHub:

# install.packages("devtools")
devtools::install_github("BDB-Genomics/EvoR")

---

🔧 Setup

To query Evo2, you need an NVIDIA BioNeMo API key. Set it as an environment variable:

Sys.setenv(NVIDIA_API_KEY = "your_nvidia_api_key_here")

---

⚡ Quick Start

Here is a complete end-to-end example workflow:

library(EvoR)

# 1. Define a DNA sequence of interest
sequence <- "ATGCGTACGTAGCTAGCTAGCTAGCTAGC"

# 2. Query Evo2 for sequence generation & scoring
response <- evo2_query(sequence, num_tokens = 10)

# 3. Extract confidence/variant scoring
scores <- evo2_get_score(response)
print("Nucleotide log-probabilities:")
print(head(scores))

# 4. Extract sequence embeddings (forward pass)
emb_response <- evo2_query_embeddings(sequence)
embeddings <- evo2_get_embeddings(emb_response)

# The result is a numeric matrix [sequence_length x 4096]
print("Embeddings Matrix Dimensions:")
print(dim(embeddings))

---

📚 References

• Evo2 Nature Paper: Arc Institute / Nature (2026)
• Source Code: BDB-Genomics/EvoR

---

Developed as part of the BDB-Genomics ecosystem.