BiologybioRxiv
Ranked by formulaVerdict: NotableScaling a context-aware genomic model for condition-specific edit design in plants
Engineering agronomically relevant traits in crops often requires changing a gene's activity in a specific tissue, developmental stage or environmental condition while preserving its roles in other contexts.
Key numbers
- 246% higher median correlations than
Caveats
- Preprint; not yet peer reviewed.
VerdictWorth a reader's time today.
Abstract
Engineering agronomically relevant traits in crops often requires changing a gene's activity in a specific tissue, developmental stage or environmental condition while preserving its roles in other contexts. Designing such selective changes requires predicting how cis-regulatory sequences are interpreted in different biological states, as the same edit can have different effects across contexts. Genomic models have made substantial progress in predicting gene expression from DNA sequence, yet sequence-only predictors do not receive an explicit description of biological states. We introduce CROP-1 (Context-aware Regulatory Output Prediction model), which combines DNA sequence with transcription factor binding motifs and measured expression to predict gene activity as a function of both sequence and regulatory state. On a held-out test set spanning six plant species, CROP-1 achieves a median per-gene Pearson correlation across contexts of 0.831, compared with 0.486 for its parameter-matched sequence-only counterpart, and 57-246% higher median correlations than three genomic foundation models fully fine-tuned on the same task. Prediction accuracy also improves with scale: increasing model capacity reduces validation error by 48%, while expanding training data to include more diverse biological contexts reduces it by 58%. We then use gradient-based sequence optimisation to translate a desired gene expression pattern into candidate edits of its cis-regulatory sequence, targeting up- or down-regulation in one context while preserving expression in others. In soybean and tomato, edit designs with 1-20 nucleotide substitutions produce strong predicted target responses and much smaller average changes across non-target contexts. By connecting scalable learning of sequence-state relationships to selective edit design, CROP-1 provides a computational foundation for controlling where, when and under which conditions endogenous crop genes are active.
The editor's rubric
- Leverage3 / 5
- A method or resource many groups across the field will adopt within a year.
- Magnitude3 / 5
- Large gain: roughly 2x, or a clear new state of the art on a hard, unsaturated problem.
- Evidence3 / 5
- Solid: multiple benchmarks or cohorts, ablations, fair baselines, released code or data.
- Novelty2 / 5
- A new combination of known ideas.
- Trajectory3 / 5
- A clear path to scale.
- Stakes2 / 5
- Benefits a professional community (practitioners, clinicians, engineers).
Editor’s rationale
Heuristic triage from title and abstract text only, not a reading of the paper. Cues found: method (we propose); breadth (many tasks); gains (relative gain, versus baseline); verification (multiple benchmarks, held-out test); scale (scalable, improves with scale).
How the score was computed
Desks place stronger verdicts first, then use this score to order work within the same verdict.
- Merit
- 5.4 / 10
- Adjusted merit
- 4.6 / 10
- Attention
- 0%
- Freshness
- 88%