Introduction

Profluent Bio has published the first demonstration of gene editing in human cells using components designed entirely by artificial intelligence. Their system, OpenCRISPR-1, was not discovered in nature but generated from scratch by large-scale protein language models trained on massive genomic data. This represents a new era where AI does not just analyze biological data but creates functional molecular machines.

Model Training and Design Process

• Dataset: The team mined over 26 terabases of genomic data, extracting more than one million CRISPR operons.
• Training: These sequences were used to train generative AI models capable of learning the “language” of CRISPR systems.
• Output: Thousands of novel protein candidates were generated. One, named OpenCRISPR-1, was selected for functional testing in human cells.

Experimental Validation

1. On-target activity:
   OpenCRISPR-1 editing efficiency matched or exceeded SpCas9, the widely used natural CRISPR system.
   In comparative assays, OpenCRISPR-1 showed equivalent gene disruption capability at target sites.
2. Off-target reduction:
   Off-target activity was reduced by approximately 95% compared to SpCas9 (0.32% vs. 6.1%).
   This addresses one of the central limitations of CRISPR technology, where unintended DNA cuts can compromise safety.
3. Base editing compatibility:
   When converted into a nickase and fused to deaminases, OpenCRISPR-1 supported efficient A-to-G base editing at rates of 35–60%, with minimal insertions or deletions.
   Both established deaminases (e.g., ABE8.20) and AI-generated enzymes (PF-DEAM-1, PF-DEAM-2) functioned effectively.
4. Guide RNA optimization:
   Profluent also trained neural networks to generate optimized single-guide RNAs (sgRNAs).
   Several of these AI-designed guides improved editing efficiency compared with standard designs.
5. Safety considerations:
   OpenCRISPR-1 lacks several known immunogenic epitopes present in SpCas9, potentially reducing immune response risk in therapeutic contexts.

Open Science Approach

Unlike proprietary editing systems, Profluent has released OpenCRISPR-1 openly, including:

• DNA sequences
• Plasmids
• Source code
• Documentation under ethical licensing terms

This decision is intended to maximize accessibility for academic and therapeutic research while encouraging community oversight.

Significance

OpenCRISPR-1 demonstrates that AI can move beyond prediction to generative design of functional biological systems. The technology shifts gene editing away from trial-and-error and toward precise, model-driven engineering. Future applications include:

• Safer and more accurate therapeutic genome editing
• Personalized gene therapy tailored to individual genomes
• Expansion into base editing, prime editing, and other precision DNA manipulation strategies

References

1. Profluent Bio. Editing the Human Genome with AI (2025). https://www.profluent.bio/media/editing-the-human-genome-with-ai
2. Nature. Generative design of CRISPR proteins with language models (2025). https://www.nature.com/articles/s41586-025-09298-z
3. CRISPR Medicine News. OpenCRISPR-1: Generative AI Meets CRISPR (2025). https://crisprmedicinenews.com/news/opencrispr-1-generative-ai-meets-crispr
4. Genetic Engineering & Biotechnology News. Profluent’s AI-Designed Gene Editor Glimpses Into a Generalizable Platform (2025). https://www.genengnews.com/topics/artificial-intelligence/profluents-ai-designed-gene-editor-glimpses-into-generalizable-platform
5. Oxford Global. Profluent Launches AI-Enabled OpenCRISPR-1 (2025). https://oxfordglobal.com/discovery-development/resources/new-trailblazer-profluent-launches-its-ai-enabled-opencrispr-1-to-edit-the-human-genome