Large-scale DNA-based phenotypic recording and deep learning enable highly accurate sequence-function mapping.

Area of Science:

• Genomics and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Predicting gene regulatory element (GRE) function is crucial but challenging due to limitations in generating large-scale sequence-function datasets.
• Existing experimental methods for GRE dataset generation are often application-specific, technically complex, or prone to errors.

Purpose of the Study:

• To introduce a widely applicable and practicable method, DNA-based phenotypic recording, for generating large-scale sequence-function datasets.
• To develop and apply a deep learning model for accurate prediction of GRE function from sequence data.

Main Methods:

• Developed DNA-based phenotypic recording using site-specific recombinase to directly record GRE effects in DNA.
• Enabled high-throughput readout of sequence and quantitative function for millions of GREs via next-generation sequencing.
• Applied a deep learning approach with ensembling and uncertainty modeling to predict ribosome binding site (RBS) function.

Main Results:

• Successfully recorded translation kinetics for over 300,000 bacterial RBSs, generating >2.7 million sequence-function pairs in a single experiment.
• The deep learning model achieved high accuracy in predicting RBS function, surpassing existing state-of-the-art methods.
• Demonstrated the efficacy of DNA-based phenotypic recording for generating comprehensive sequence-function datasets.

Conclusions:

• DNA-based phenotypic recording provides a scalable and practical solution for generating large sequence-function datasets.
• The combination of this recording method with advanced deep learning significantly enhances the ability to predict biological function from genetic sequences.
• This integrated approach represents a substantial advancement in understanding and predicting gene regulation.