Preface

DeepSeek Harness (DSH) uses a plugin-based architecture, and its ecosystem is maintained by the community. When building bioinformatics agents, domain-specific design tools are required for particular areas such as gene editing. dsh-bio-graft is part of the dsh-bio family, maintained by moonbowterfly, and focuses on CRISPR/gene-editing design workflows.

Plugin Positioning

dsh-bio-graft is a specialized gene-editing design plugin for DSH. Through its graft_* semantic toolset, it provides end-to-end capabilities ranging from sgRNA candidate enumeration, cut-site calculation, and score vector retention to off-target scanning, base-editing design, and validation plan generation.

Core Capabilities

The plugin provides 11 semantic tools, with core capabilities as follows:

  • NucleaseProfile Registry: Built-in editor geometry evidence grading (verified/pam_source), distinguishing unverified statuses (e.g., Cas12b/Cas13a).
  • sgRNA Candidate Enumeration and Scoring: Uses graft_design to scan PAMs and generate candidate lists. The tool retains the full scoring vector (GC, homopolymers, seed sequence, etc.) rather than generating a single composite score; cut-site calculation supports IUPAC awareness.
  • Declarative Ranking: Uses graft_rank to rank candidates using Pareto or Lexicographic ordering, outputting policy_id and policy_digest, with rejected candidates accompanied by explicit reasons.
  • Off-Target Scanning: Invokes Cas-OFFinder via graft_offtarget for structured scanning. The tool returns interpretation_boundary (a machine-readable boundary declaration); when zero hits are found, it lists common false-negative causes (e.g., genome version mismatch, N gaps).
  • Base-Editing Design: Uses graft_base_edit to design CBE/ABE, supporting in-window base editing, bystander effects, and codon consequence analysis (synonymous/missense, etc.).
  • Multi-Strategy Evaluation: Uses graft_strategy to evaluate multi-guide strategies (e.g., deletion_pair, paired_nickase), calculating predicted deletion intervals and pairwise off-target combinations.
  • Validation Plan Generation: Uses graft_validation_plan to generate validation metrics including numerator/denominator/assay, with support for plant-specific requirements.
  • EditPlan Ledger Management: Manages auditable .editplan.json ledges via graft_plan_save and graft_plan_load, supporting historical recall.

Installation and Enablement

As a DSH plugin, it can be installed with the following command:

dsh plugin --profile web add @dsh-bio/dsh-bio-graft

After installation, the tool registry and Genie settings panel will sync automatically, with no additional configuration required.

Typical Usage

The plugin supports the following typical operations:

  1. Scan and Enumerate: Use graft_design to scan PAMs and generate candidates.
  2. Rank and Filter: Use graft_rank to perform declarative ranking of candidates.
  3. Off-Target Detection: Use graft_offtarget to perform off-target detection.
  4. Base Editing: Use graft_base_edit to design base editing.
  5. Strategy Evaluation: Use graft_strategy to evaluate multi-guide strategies.
  6. Validation Plan: Use graft_validation_plan to generate a validation plan.

Notes

  • Runtime Dependencies:
    • Requires Python >= 3.10 (the dsh-bio-genie environment can be reused).
    • Off-target scanning depends on Cas-OFFinder (BSD-3), which must be placed manually or fetched automatically via graft_backend_status action=ensure.
    • Requires the OpenCL runtime (included with GPU drivers; additional installation is required for CPU-only runs).
  • Safety Grading: The plugin includes Level 0-3 safety gating, with different design restrictions for human somatic cells (Level 1), clinical decision-making (Level 2), and heritable editing (Level 3).
  • License Policy: MIT licensed, with no license-incompatible code bundled (e.g., GPL-3+ FlashFry).

Conclusion

dsh-bio-graft provides the DSH ecosystem with a complete gene-editing workflow from design to validation. By retaining scoring vectors and auditable EditPlans, it is suitable for rigorous research and development scenarios that require strict experimental controls.