Sequence-to-function deep learning frameworks for engineered riboregulators

Abstract

While synthetic biology has revolutionized our approaches to medicine, agriculture, and energy, the design of completely novel biological circuit components beyond naturally-derived templates remains challenging due to poorly understood design rules. Toehold switches, which are programmable nucleic acid sensors, face an analogous design bottleneck; our limited understanding of how sequence impacts functionality often necessitates expensive, time-consuming screens to identify effective switches. Here, we introduce Sequence-based Toehold Optimization and Redesign Model (STORM) and Nucleic-Acid Speech (NuSpeak), two orthogonal and synergistic deep learning architectures to characterize and optimize toeholds. Applying techniques from computer vision and natural language processing, we ‘un-box’ our models using convolutional filters, attention maps, and in silico mutagenesis. Through transfer-learning, we redesign sub-optimal toehold sensors, even with sparse training data, experimentally validating their improved performance. This work provides sequence-to-function deep learning frameworks for toehold selection and design, augmenting our ability to construct potent biological circuit components and precision diagnostics.

A deep learning approach to programmable RNA switches

“…Moreover, the validation of our deep-learning models on an external previously characterized dataset, as well as the holdout prediction of every individual viral genome in our dataset, further demonstrated the robust biological generalization of our models. Collaborative efforts by Valeri et al 17 also extended our work, with the implementation of a natural language modeling approach and the de-novo design and testing of toehold switches using deep-learning models.…”

“…To achieve our goal in collaboration with Valeri et al 17 , we first expand the size of available toehold datasets using a highthroughput DNA synthesis and sequencing pipeline to characterize over 10 5 toehold switches. We then use this comprehensive dataset to demonstrate that deep neural networks trained directly on switch RNA sequences can outperform rational thermodynamic and kinetic analyses to predict toehold-switch function.…”

MoiRNAiFold: a novel tool for complex in silico RNA design

“…Very recently, two publications reported Deep Learning approaches applied to thousands of toehold switches in order to infer characteristics that render some of them extremely efficient in vitro ( 39 , 40 ). This approach is orthogonal to our approach, however, they require extensive experimental validations that are costly and time consuming, and would have to be carried out for any new RNA constructs that can be devised in the future.…”

Massively parallel characterization of engineered transcript isoforms using direct RNA sequencing

“…By developing a new nanopore-based dRNA-seq characterization approach (Figure 1), we were able to simultaneously measure the termination efficiency of an entire mixed pool of 1183 unique transcriptional valves as well as provide nucleotide resolution insight into precisely where termination occurred for each (Figure 3). Such detail is lost with more typical fluorescence-based assays 6,11 , but is essential for developing the low-level biophysical or machine learning based models of genetic parts that are essential for predictive biodesign workflows [44][45][46][47] . While rich, highcontent characterization data can normally only be produced for a small set of samples 8,13,36,48 , the approach presented here removes this limitation, allowing us to more systematically explore the genetic design space of a large pooled library and glean several design principles.…”