henhao
Sign in
BioinformaticsbioRxiv
Preprint

Inference of CRISPR Edits from Sanger Trace Data

Tim Hsiau, David Conant, Nicholas Rossi, Travis J. MauresiD, Kelsey Waite, Joyce L. YangiD, Sahil JoshiiD, Reed KelsoiD, Kevin HoldeniD, Brittany L. Enzmann, Rich StoneriD

Abstract

Efficient precision genome editing requires a quick, quantitative, and inexpensive assay of editing outcomes. Here we present ICE (Inference of CRISPR Edits), which enables robust analysis of CRISPR edits using Sanger data. ICE proposes potential outcomes for editing with guide RNAs (gRNAs) and then determines which are supported by the data via regression. Additionally, we develop a score called ICE-D (Discordance) that can provide information on large or unexpected edits. We empirically confirm through over 1,800 edits that the ICE algorithm is robust, reproducible, and can analyze CRISPR experiments within days after transfection. We also confirm that ICE strongly correlates with next-generation sequencing of amplicons (Amp-Seq). The ICE tool is free to use and offers several improvements over current analysis tools. For instance, ICE can analyze individual experiments as well as multiple experiments simultaneously (batch analysis). ICE can also detect a wider variety of outcomes, including multi-guide edits (multiple gRNAs per target) and edits resulting from homology-directed repair (HDR), such as knock-ins and base edits. ICE is a reliable analysis tool that can significantly expedite CRISPR editing workflows. It is available online at ice.synthego.com, and the source code is at github.com/synthego-open/ice

PDFDOICode
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
302
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 3 (10 Aug 2019) · CC BY-NC-ND 4.0

ICE regresses simulated edit traces onto Sanger reads to quantify indels, multi-guide and knock-in edits without tuning, correlating with amplicon sequencing (r2 = 0.93) and TIDE (r2 = 0.99).

Checking CRISPR editing outcomes needs an assay that is fast, quantitative and cheap: amplicon next-generation sequencing (Amp-Seq) is sensitive but slower and costlier per sample unless many samples are batched, while the Sanger-based TIDE needs manual parameter tuning, lacks a batch mode and handles only single-guide edits. ICE re-implements and extends TIDE with automatic trace alignment, Lasso regression, and edit proposals for multi-guide, knock-in and base edits.

Key results

  1. ShownICE indel estimates from Sanger traces closely track amplicon sequencing (Amp-Seq), the reference method.

    Correlation across indel sizes in 92 samples sequenced by the MGH CRISPR sequencing service: overall r2 = 0.93, comparing averaged Sanger replicates with one Amp-Seq replicate; appendix examples show ICE-predicted sequences matching Amp-Seq contigs. · Results; Sanger & Next-Generation Sequencing; Appendix A: Examples of sequences predicted by…

  2. ShownICE reproduces TIDE's editing-efficiency estimates while requiring no manual parameter tuning.

    342 single-guide samples with ICE r2 > 0.95, chosen from 1805 to span 0-95% editing: ICE vs TIDE r2 = 0.99, ICE-D vs TIDE r2 = 0.97. TIDE, run on its website in August 2018, sometimes needed parameter tuning after its default setup failed. · Results; Comparison of ICE and TIDE

  3. SuggestedICE also quantifies knock-in (homology-directed repair, HDR) and single-base edits, but this text gives no correlation values for these tests.

    45 further Amp-Seq samples at 15 cut sites, with donor inserts of 0-36 bp, showed high ICE vs Amp-Seq agreement for HDR and non-homologous end joining outcomes; mixes of rs2072579 amplicons from PGP1 (G/G) and HEK293 (C/C) cells at 5% to 95% mimicked base editing, and predictions tracked the expected fractions closely. · Results; Simulated Homology-Directed Repair and Base…; Editing of Cell Cultures

  4. SuggestedThe ICE-D discordance score detects editing that the supplied edit proposals miss, such as large deletions made by guides the user did not list.

    Calibrated by least squares between indel percentage and mean downstream discordant signal over 1805 edits. When multi-guide samples were re-analysed with one guide, ICE fits were poor (r2 ≤ 0.95) and can underestimate indels, yet ICE-D still detected editing; no detection rate is given in this text. · ICE-D Algorithm; Results

  5. SuggestedLasso (L1) regularization curbs spurious indel calls, which the authors attribute to the regression overfitting Sanger background noise.

    Many indels ICE reported were absent from Amp-Seq; an L1 parameter of 0.8 reduced these false positives, tightened the correlation with Amp-Seq and gave indel profiles closer to Amp-Seq. Counts and correlation values are not in this text. · Results; ICE Algorithm

Methods, data and reproducibility

ICE re-implements TIDE in Python: it aligns control and edited Sanger traces on a high-quality window upstream of the cut site, simulates traces for candidate edits (by default deletions up to 30 bp and insertions up to 14 bp, plus multi-guide, HDR and base-edit proposals) and fits them to the edited trace by Lasso regression; ICE-D adds a calibrated discordance score.

  • Batch of CRISPR edits (Synthego sgRNA-Cas9 ribonucleoproteins in various cell lines) analysed with ICE for robustness testing and ICE-D calibration · n = 1805 edits
  • Multi-guide set: 32 genes, each targeted by three sgRNAs individually and together, with three to four Sanger replicates per edit and one replicate sent for Amp-Seq · n = 128 samples; one gene (four samples) dropped after failed Sanger runs
  • Amp-Seq validation samples (MGH CRISPR sequencing service) · n = 92 samples
  • TIDE comparison: single-guide samples with ICE r2 > 0.95 spanning a range of indel percentages · n = 342 of 1805 samples
  • Knock-in edits in HEK293 cells with ssDNA donors (+0, +14 or +36 bp inserts, 40 bp homology arms), checked by Amp-Seq · human · n = 45 samples at 15 cut sites
  • Simulated base-editing mixes of rs2072579 amplicons from PGP1 iPSCs (G/G) and HEK293 cells (C/C), 5% to 95% PGP1 · human
Methods & tools
ICE, ICE-D, TIDE, Amp-Seq, MGH NGS data pipeline, Lasso regression, Sanger sequencing, Synthego sgRNAs, Lonza Nucleofector, QuickExtract DNA Extraction Solution, Fragment Analyzer (AATI)
Reagents & environment
Python re-implementation of TIDE; a Docker container is published on Docker Hub as synthego/ice. No package versions are stated.
Experimental conditions
About four seconds per sample on average, single-threaded on a laptop (MacBook Pro 2017).
Reproducibility gaps
The source code and webtool links are missing from this text (the availability statements end in 'at'), and no availability statement or accessions are given for the Sanger traces or Amp-Seq data.
Limitations (6)
  • ICE and TIDE assume Sanger peak height is proportional to base molarity with one coefficient for all bases, yet peak height and phasing depend on sequence context; this may matter most for HDR and base editing, which rest on single positions. · stated by the authors
  • Multi-guide edit proposals can include impossible events, such as two guides cutting at adjacent positions and dropping the base between them. · stated by the authors
  • Amp-Seq remains more sensitive and more quantitative than Sanger-based analysis. · stated by the authors
  • When the supplied guides do not cover all outcomes, ICE (like TIDE) fits poorly and may underestimate indels; ICE-D is offered as a safeguard. · stated by the authors
  • The TIDE comparison used only samples with a high ICE fit (r2 > 0.95), which favours agreement, and the HDR, base-editing, ICE-D and regularization results come without numbers in this text. · noted in this summary
  • The Amp-Seq comparison reports a single overall r2 across indel sizes rather than per-sample error in editing efficiency, and the lowest edit frequency ICE can reliably detect is not quantified. · noted in this summary

Who should read:Read this if you check CRISPR knockout, multi-guide, knock-in or base-editing outcomes by Sanger sequencing and want an automated, tuning-free alternative to TIDE or a cheaper stand-in for amplicon sequencing.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISSuryawanshi '25Labun '19Kluesner '18Clement '19Hwang '18Sanvicente- '23
66% semantic matchBioinformatics
Gajendra W. Suryawanshi·bioRxiv·Jan 23, 2025