PPubMed11 Jan 2024Cited 26×

Improved prediction of bacterial CRISPRi guide efficiency from depletion screens through mixed-effect machine learning and data integration

Yanying Yu, Sandra Gawlitt, Lisa Barros de Andrade e Sousa, Erinc Merdivan, Marie Piraud, Chase L. Beisel, Lars Barquist

Abstract

'Erinc Merdivan' 'Marie Piraud' 'Chase L. Beisel' 'Lars Barquist'] CRISPR interference (CRISPRi) is the leading technique to silence gene expression in bacteria; however, design rules remain poorly defined. We develop a best-in-class prediction algorithm for guide silencing efficiency by systematically investigating factors influencing guide depletion in genome-wide essentiality screens, with the surprising discovery that gene-specific features substantially impact prediction. We develop a mixed-effect random forest regression model that provides better estimates of guide efficiency. We further apply methods from explainable AI to extract interpretable design rules from the model. This study provides a blueprint for predictive models for CRISPR technologies where only indirect measurements of guide activity are available. Supplementary Information The online version contains supplementary material available at 10.1186/s13059-023-03153-y.

A figure from Improved prediction of bacterial CRISPRi guide efficiency from depletion screens through mixed-effect machine learning and data integration — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

A mixed‑effect random forest model integrating multiple CRISPRi screens predicts bacterial guide efficiency better than previous methods, revealing strong gene‑specific effects.

What’s new

i.Developed a mixed‑effect random forest (MERF) model that separates guide and gene effects for CRISPRi efficiency prediction
ii.Identified gene‑specific features as major determinants of guide depletion
iii.Integrated three genome‑wide CRISPRi essentiality screens to improve model performance
iv.Created the CIAO web server for easy guide‑efficiency prediction
v.Validated the model with a high‑throughput saturating screen of purine‑biosynthesis genes

How they did it

The authors compiled 8099 gRNAs from three E. coli CRISPRi essentiality screens and engineered 138 sequence, thermodynamic, genomic, and transcriptomic features. They applied automated machine learning (auto‑sklearn) and a mixed‑effect random forest regression (MERF) with ten‑fold cross‑validation, using SHAP explainable AI to extract design rules. Model performance was further validated with a saturating screen of 750 gRNAs targeting purine‑biosynthesis genes.

What they found

MERF trained on three datasets achieved a median Spearman correlation of 0.396, outperforming the retrained Pasteur model (ρ = 0.366, p = 0.014)
Deep learning models did not improve prediction performance, likely due to limited training data
The saturating purine screen used 750 gRNAs (35–223 per gene) across nine genes, confirming MERF’s superior accuracy

Caveats

✕Deep learning approaches failed to surpass MERF, possibly because of limited training data
✕Low‑throughput validation data were too sparse to draw strong conclusions

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Similar papers in all four archives

The same idea searched in arXiv, PubMed, bioRxiv and chemRxiv at once, so a method can lead you to where it is being applied.

AarXiv4 found

PPubMed3 found

BbioRxiv4 found

CchemRxiv3 found

Searching the other archives…