Comment from A. Daneshpajouh

Executive summary. This comment asks the Agency to require four additive, quantitative reporting elements layered onto the existing draft guidance, each anchored to a specific guidance section and each computable today on public data alone: 1. (Sections VI-VII, off-target nomination completeness.) A sponsor relying on an in silico off-target candidate list should disclose the completeness of that list against three explicit, computable channels - reference-only, patient short variants, and patient structural-variant junctions - rather than leaving the list's completeness unstated. 2. (Section VIII, chromosomal-integrity / loss-of-genome-integrity analysis.) A sponsor relying on a short-read structural-variant caller for large-indel or translocation detection should report a distribution-free, caller-specific certified miss-rate floor against an orthogonal high-fidelity (e.g., long-read) truth set, because short-read SV callers carry large, caller-specific blind spots that a "sensitive and quantitative" claim (guidance's own language) does not, by itself, rule out. 3. (Sections IV-VII, sequencing-depth / sample-size adequacy.) Any zero-observed-failure calibration or validation claim should distinguish an existence floor (the sample size at which a bound is merely defined) from the much larger certification floor (the sample size actually needed to certify the stated miss-rate), because the two are numerically far apart and are conflated in current practice. 4. (Submission of study reports, bioinformatics reproducibility.) Every certified or reported number should be traceable to a hash-bound, independently re-derivable artifact and pipeline, as a concrete operationalization of the guidance's reproducibility expectations. Two worked examples (an approved ex vivo product and an investigational in vivo product) show the off-target completeness gap is measurable today on public reference data, and that for at least one documented site the ancestry-conditional coverage gap spans two orders of magnitude (8.9% vs 0.029% carrier probability, AFR vs NFE, at a single created candidate). An independent public-data benchmark shows that a widely used short-read structural-variant caller, evaluated against a long-read gold-standard truth set, has a certified deletion miss-rate of at least 70.7% (one-sided 95% lower bound) on the benchmarked class - a concrete instance of the detection-floor gap that Section VIII's "sensitive and quantitative" language does not itself quantify. ------------------------------------------------------------------------ Status of this comment (read first) This is a methodology proposal, not a safety alert. Its recommendations are that, where a sponsor's genome-editing safety assessment relies on (a) an in silico nominated off-target candidate list, (b) a short-read NGS-based structural-variant/translocation assay, or (c) a zero-failure validation claim at any stage, the sponsor report the completeness or certified detection floor of that step against a small, explicit, computable metric, with residuals stated as labeled bounds rather than left implicit. The draft guidance (Docket FDA-2026-D-1255, Federal Register Doc. 2026-07285, published April 15, 2026; comment period closes July 14, 2026) directs sponsors to "provide data generated in-house... and/or relevant information from peer-reviewed publications, to support the adequacy and the sensitivity of sequencing depth and to support their strategy to detect low frequency off-target editing events" (Section IV), to perform "sensitive and quantitative NGS-based assessment of genomic integrity" for chromosomal-integrity analysis (Section VIII), and to account for human genetic variation in off-target nomination (Section VII) - but in each case it does not specify a statistical form for the completeness, sensitivity, or certified-floor claim. This comment offers four concrete, computable instruments to fill that gap, each mapped to its named guidance section. Nothing here asserts that any worked-example site is an active, edited, or harmful off-target, nor that any specific sponsor's assay is inadequate. Every activity statement is explicitly separated into certified, priced (computable but requiring patient/assay data), and wet-lab-gated (unknown until measured). Each quantitative claim is cited to the persisted result file that holds it (Section 8), and the per-ancestry coverage-gap masses, the SV-junction counts, the created-site carrier figures, and the SV-caller detection-floor figures each reproduce to the digit against the file named for them. ------------------------------------------------------------------------ Full 23-page comment is attached as FDA_COMMENT_FINAL.pdf. Affiliation: Simon Fraser University's Computational Biology Lab, under the supervision of Dr. Kay C. Wiese. Conflict of interest: The author declares no competing interests.