How Helix Studio thinks.

A transparent look at the algorithms and references backing Helix Studio's design, evidence, and analysis stack.

Guide scoring

Helix Studio blends internal cut-site regression models with open benchmarks (Doench 2016, Moreno-Mateos) and recalibrates monthly against pilot data.

Output: normalized 0–1 score + constraint explanations

Off-target risk

GPU-accelerated alignment surfaces top N mismatches and bulges with CFD-style weighting. CLI export includes JSON for downstream filters.

Supports HG19, HG38, CHM13, and custom references

Prime Editing efficiency

pegRNA scoring uses the Anzalone feature set plus Helix's own indel/RTT regressors. Evidence generation runs multiple RTT/PBS permutations to suggest safer edits.

Outputs efficiency percentile + recommended adjustments

Indel spectrum

Amplicon reads feed into a transformer alignment head trained on >5M labeled events. Results highlight frame shifts, splice hits, and microhomology events.

Benchmarked vs. synthetic chr20 data + GIAB references