Each DNA molecule embodying this architecture contains information about both the GRE sequence and a measure of its function (i.e. Thus, a physical link between GRE and the recombinase substrate is established, and the latter serves as stable, heritable record of the GRE’s effect on gene expression. Its core innovation is a three-component genetic architecture that combines on the same DNA molecule the gene of a site-specific DNA recombinase, a GRE controlling its expression, and the recombinase substrate. Here, we introduce a method that relies on DNA-based phenotypic recording to address the limitations enumerated above. Therefore, the need for widely applicable, technically simple and yet accurate high-throughput approaches to ascribe functional (or phenotypic) readouts to genetic sequences persists. RNA sequencing techniques avoid some of these limitations but are restricted to transcriptional effects and can be greatly biased due to variability in reverse transcription, barcode-induced bias, and DNA amplification efficiencies 26, 27. However, these approaches either require elaborate sample processing procedures, which are prone to introduce bias, or are restricted to specific functional readouts. In a particularly noteworthy recent study, Yus and coworkers have used dam methylase to facilitate a functional readout quantifiable by NGS with high throughput 24. Furthermore, ribosome loading 23, DNA methylation 7, 24, and enrichment by growth selection 25 have been suggested in combination with NGS as alternative approaches. This introduces errors and limits data quality 21, 22 impairing prediction accuracy. In previous efforts to alleviate this experimental bottleneck 3, 18, 19, 20, 21, the functional readout is performed in a separate technical step, and retroactively mapped back to the corresponding sequence by statistical inference. These methods promise to be able to model sequence–function dependencies with minimal prior assumptions, provided that large experimental training data sets that link sequence to quantitative measure of function 16, 17 are available.Īlthough next-generation sequencing (NGS) allows obtaining sequence information at extremely large scale, our ability to assign a quantitative functional readout to each sequence has not kept pace. Deep learning maximizes the benefit of big data collection owing to its ability to capture complex, non-linear dependencies and to its computational scalability 9, which led to several successful applications in computational biology, from genomics to proteomics 10, 11, 12, 13, 14, 15. At the same time, novel methods are required that identify statistical patterns and dependencies in the resulting data sets to generate models that accurately predict the properties of untested sequences. Therefore, innovative high-throughput (HTP) approaches are required that allow to collect a quantitative functional readout for large numbers of genetic sequences 7, 8.
As the number of possible sequences scales exponentially with their length, the theoretical sequence space cannot be exhaustively explored by experiments, even for small GREs 5, 6, 7. Despite this progress, the relationship between a genetic sequence and its functional properties is poorly understood, and thus the question what to write remains largely unanswered 3, 4. Recent progress in DNA sequencing and synthesis has facilitated reading and (re-)writing of the genetic makeup of biological systems on a massive scale 1, 2. DNA-based phenotypic recording combined with deep learning represents a major advance in our ability to predict function from genetic sequence.
Further, we introduce a deep learning approach employing ensembling and uncertainty modelling that predicts RBS function with high accuracy, outperforming state-of-the-art methods. We record translation kinetics of over 300,000 bacterial ribosome binding sites (RBSs) in >2.7 million sequence-function pairs in a single experiment. We use a site-specific recombinase to directly record a GRE’s effect in DNA, enabling readout of both sequence and quantitative function for extremely large GRE-sets via next-generation sequencing. Here, we introduce DNA-based phenotypic recording as a widely applicable, practicable approach to generate large-scale sequence-function datasets. However, experimental methods to generate such datasets are either application-specific or technically complex and error-prone. Machine learning may address this, but requires large datasets linking GREs to their quantitative function. Predicting effects of gene regulatory elements (GREs) is a longstanding challenge in biology.
0 kommentar(er)
