A deep-learning predictive model for Cas12a activity in wheat

CRISPR technology has created targeted DNA mutations across the tree of life. Target sites are defined by RNA molecules that largely determine the efficiency of an experiment. Several predictive models are available to select the most efficient RNAs. Unfortunately, these models are not built with plant data and have limited predictive power in plant systems. To overcome this limitation, we established a high-throughput arrayed wheat protoplast transfection method and measured the editing efficiency of ~1,000 unique LbCas12a crRNAs at on- and off-target sites. We trained different models by testing 40 deep learning architectures and found that a convolutional neural network model called 2CaP-5 displayed the best performance. While 2CaP-5 had moderate to strong predictive power in protoplasts, it did not generalize to Agrobacterium-mediated transformation. Despite this limitation, we demonstrate that 2CaP-5 can be re-trained with as few as 500 crRNAs. The use of such a small training set is advantageous for low-throughput experimental systems such as typically found in transgenic plant systems. We also used ~2,000 off-targets in the wheat subgenomes to investigate Cas12a specificity. Our dataset confirms previous observations regarding off-targets, specifically that single mismatches reduce indel efficiency provided they are adjacent to the PAM and that off-target activity can be largely prevented by three mismatches. Altogether, our results show that the 2CaP-5 model has moderate to strong predictive power for Cas12a crRNAs in wheat protoplasts and will serve as a useful foundation for predictive tools in other plant species.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-20
DOI
https://doi.org/10.64898/2026.11.18.767513
Primary Topic
CRISPR and Genetic Engineering
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A deep-learning predictive model for Cas12a activity in wheat

Alexandra Peña Fernández, Christophe Gaillochet, Jonas De Saeger, Myriam Shafie et al.
2 links
bioRxiv (Cold Spring Harbor Laboratory)
CRISPR and Genetic Engineering
preprint

A deep-learning predictive model for Cas12a activity in wheat

Alexandra Peña Fernández, Christophe Gaillochet, Jonas De Saeger, Myriam Shafie, Laurens Pauwels, Ward Develtere, Frank Meulewaeter, Pieter Wytynck, Lieven Sterck, Els Van Lerberge, Ive De Smet, Ellen Slaman, Debbie Rombaut, David De Vleesschauwer, Marnik Vuylsteke, Dirk Inze, Denys Marushchak, Thomas Benjamin Jacobs, Tingting Zhu
preprint en

Abstract

CRISPR technology has created targeted DNA mutations across the tree of life. Target sites are defined by RNA molecules that largely determine the efficiency of an experiment. Several predictive models are available to select the most efficient RNAs. Unfortunately, these models are not built with plant data and have limited predictive power in plant systems. To overcome this limitation, we established a high-throughput arrayed wheat protoplast transfection method and measured the editing efficiency of ~1,000 unique LbCas12a crRNAs at on- and off-target sites. We trained different models by testing 40 deep learning architectures and found that a convolutional neural network model called 2CaP-5 displayed the best performance. While 2CaP-5 had moderate to strong predictive power in protoplasts, it did not generalize to Agrobacterium-mediated transformation. Despite this limitation, we demonstrate that 2CaP-5 can be re-trained with as few as 500 crRNAs. The use of such a small training set is advantageous for low-throughput experimental systems such as typically found in transgenic plant systems. We also used ~2,000 off-targets in the wheat subgenomes to investigate Cas12a specificity. Our dataset confirms previous observations regarding off-targets, specifically that single mismatches reduce indel efficiency provided they are adjacent to the PAM and that off-target activity can be largely prevented by three mismatches. Altogether, our results show that the 2CaP-5 model has moderate to strong predictive power for Cas12a crRNAs in wheat protoplasts and will serve as a useful foundation for predictive tools in other plant species.

bioRxiv (Cold Spring Harbor Laboratory)
VIB-UGent Center for Plant Systems Biology (BE)
Agentschap Innoveren en Ondernemen, China Scholarship Council, Vlaamse regering
Zero hunger
CRISPR and Genetic Engineering
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon