CRISPR-GPT for Agentic Automation of Gene-editing Experiments

TL;DR

CRISPR-GPT presents an AI-driven design agent that combines targeted domain knowledge with tool integration to automate CRISPR gene-editing design tasks, addressing the limitations of general LLMs in biology. The system uses a modular, state-machine workflow with four core components to manage 22 tasks across four meta-tasks, enabling end-to-end planning from CRISPR system selection to validation. Expert evaluation and a real-world A375 knockout demonstration show superior accuracy and completeness compared with general LLMs, while Safety and Ethics sections outline safeguards for heritable editing and data privacy. The work signals a path toward responsible, scalable, AI-assisted design of gene-editing workflows and hints at future integration with automated laboratory platforms.

Abstract

The introduction of genome engineering technology has transformed biomedical research, making it possible to make precise changes to genetic information. However, creating an efficient gene-editing system requires a deep understanding of CRISPR technology, and the complex experimental systems under investigation. While Large Language Models (LLMs) have shown promise in various tasks, they often lack specific knowledge and struggle to accurately solve biological design problems. In this work, we introduce CRISPR-GPT, an LLM agent augmented with domain knowledge and external tools to automate and enhance the design process of CRISPR-based gene-editing experiments. CRISPR-GPT leverages the reasoning ability of LLMs to facilitate the process of selecting CRISPR systems, designing guide RNAs, recommending cellular delivery methods, drafting protocols, and designing validation experiments to confirm editing outcomes. We showcase the potential of CRISPR-GPT for assisting non-expert researchers with gene-editing experiments from scratch and validate the agent's effectiveness in a real-world use case. Furthermore, we explore the ethical and regulatory considerations associated with automated gene-editing design, highlighting the need for responsible and transparent use of these tools. Our work aims to bridge the gap between beginner biological researchers and CRISPR genome engineering techniques, and demonstrate the potential of LLM agents in facilitating complex biological discovery tasks. The published version of this draft is available at https://www.nature.com/articles/s41551-025-01463-z.

Figures (8)

• Figure 1: Overview of CRISPR-GPT Agent. The CRISPR-GPT is built upon an LLM-powered design and planning engine (left), which helps to complete 4 core meta-tasks (top right), as well as other auxiliary functions (freestyle Q&A, off-target prediction). CRISPR-GPT has integrated a set of useful skills and toolkits (bottom right) that the LLM agent would call when needed to facilitate human users across the different tasks and subtasks. Figure created with BioRender.com.
• Figure 2: Components of CRISPR-GPT enable human-AI collaboration to automate gene-editing experimental designs across complex tasks.LLM Planner is responsible for configuring tasks based on the user’s needs (4 predefined meta-tasks or LLM planned chain of tasks). Tool Provider connects the system to external APIs, tools, libraries, and documents. Task Executor is implemented as a state machine, responsible for providing instructions and feedback, receiving input from LLM Agent, and calling APIs via Tool Provider. LLM Agent is responsible for interacting with the task executor on behalf of the user, where the user can monitor the process and provide correction to the LLM agent if the generated content.
• Figure 3: Task decomposition process and state machine implementation algorithm.(Left) Task decomposition; The LLMs can automatically perform task decomposition based on the user’s request, the descriptions of the current supported tasks and the dependencies, and the internal knowledge inside the LLMs. The state machines of the selected tasks are chained together to fulfill the user’s request. (Right) State machines & LLM Agent;State machines are the core of the Task Executor, where each state is responsible for one round of interaction with the user. The instruction is provided to the user first with sufficient information for the current decision-making step and the required inputs. After receiving the response from the user, it provides output and feedback, where APIs (e.g. program execution/web search/database retrieval) are potentially called during the execution of the state. Afterward, the state machine transits to the next state. LLM Agent generates responses to every step of the state machine on behalf of the user. The user monitors the whole process and provides corrections if the generated content is wrong or overrides the LLM Agent and manually interacts with the Task Executor.
• Figure 4: Overview of CRISPR-GPT’s interactive modules for gene-editing experimental design. (A) schematics illustrating the functionalities of the three modules within CRISPR-GPT, accompanied by examples of their applications. (B) Web-interface of CRISPR-GPT, note No.1-4 is the “Meta Mode”, No.5 is the “Auto Mode”, No.6 is Off-target-prediction function and “Q: prompt would trigger the Q&A mode”.
• Figure 5: Example workflows outlining the general tasks involved in gene-editing experimental designs as facilitated by CRISPR-GPT.