bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.64898/2026.01.22.701115

Machine-learning-driven prediction and design of intrinsic transcription terminators

Kundlatsch, G. E.·Neto, A. P. d. S.·de Paiva, G. B.·Rech, E.·Duarte, L. T.·Pedrolli, D. B.
View original source ↗Export metadata ↓

Abstract

Intrinsic transcription terminators are biological parts critical for controlling gene expression in natural genomes and are fundamental to the modularity and predictability of synthetic gene circuits. Despite their simplicity of structure and function, we have not yet been able to rationally engineer synthetic terminators with a pre-defined strength, nor to accurately predict their strength from sequence. Here, we leveraged a curated library of bacterial terminators to train a data-driven predictive model, and, building on this surrogate, developed open-source software tools for predicting terminator performance and designing new intrinsic terminator sequences. Model interpretability analysis indicates that U-tract features emphasize a distal region longer than previously anticipated and that the initial hairpin GC content influence extends beyond the reported range. Using the final trained model, we implemented two software tools. The Terminator Strength Predictor (TerSP) computes the full feature representation directly from an input sequence and outputs a quantitative strength prediction together with a binary strong/weak classification. We validated TerSP using experimentally characterized terminators from bacteria other than E. coli. The Terminator Factory (TerFac) implements a surrogate-based optimization framework for target-driven terminator design under user-defined strength and length constraints. Using TerFac, we enumerated length-specific sets of maximally strong terminators, designed optimized synthetic terminators, and optimized a wild-type terminator. The designed terminators were validated in vivo in E. coli and in vitro, using a newly developed assay based on fluorescent RNA aptamers. The TerFac-designed terminators showed the expected strength, and the strongest one outperformed the best reference terminator in the training dataset, both in vivo and in vitro. These results indicate that the model captured sequence-to-function rules that are informative both for forward prediction (TerSP) and for the design of terminators with defined strength (TerFac).

Explore related subjects

synthetic biologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Kundlatsch, G. E., Neto, A. P. d. S., de Paiva, G. B., Rech, E., Duarte, L. T., Pedrolli, D. B.. 2026-01-22. Machine-learning-driven prediction and design of intrinsic transcription terminators. https://doi.org/10.64898/2026.01.22.701115

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Coupling a developmental promoter to CRISPR interference for Wnt pathway regulation in human pluripotent stem cells

While directed differentiation of human pluripotent stem cells commonly relies on the timed delivery of extracellular factors, these uniform treatments often yield heterogeneous responses across cell populations. Linking intracellular gene regulation directly to an emerging developmental state offers a complementary strategy to coordinate these differentiation signals from within the cell. Here, we explored this approach by coupling a T/Brachyury promoter to CRISPR interference targeting CTNNB1, which encodes the canonical Wnt signaling mediator {beta}-catenin. A T-promoter EGFP reporter line exhibited transiently increased activity during early differentiation, supporting the use of this promoter as a developmentally responsive input. We then combined the promoter with dCas9-KRAB and a CTNNB1-targeting guide RNA. Cas9-mediated integration was accompanied by indels at the CTNNB1 target site, whereas a Cas12a-mediated integration strategy yielded clones with no indels detected by ICE analysis. During differentiation, the selected T-dCas9 CTNNB1 clone showed reduced CTNNB1 expression and attenuated induction of Wnt associated genes. Together, these findings provide a proof of concept for combining a developmental promoter with a programmable intracellular regulator and identify a strategy for separating circuit integration from unintended target-site editing. This modular approach provides a foundation for developing genetic interventions whose expression is linked to developmental state.

synthetic biology↗
bioRxiv2026

Thermodynamic, Electrochemical and Practical Constraints on Electromicrobial Formate Assimilation

Electromicrobial production (EMP) technologies aim to combine renewable electricity, CO2, and engineered microbes to make energy-dense molecules at efficiencies exceeding photosynthesis. CO2 can be electrochemically reduced to formate, which is far easier to handle at the bench than H2 or an electrode, but formate carries only two electrons per carbon against the six in a biofuel. The remaining electrons must come from oxidizing additional formate, from H2 oxidation, or from extracellular electron uptake (EEU), and no rigorous comparison of these options coupled to the choice of carbon assimilation pathway currently exists. We calculate upper-limit efficiencies for butanol production by six carbon assimilation pathways, each paired with all three electron delivery mechanisms, using electrochemical parameters drawn from a survey of the recent literature. Electrical to butanol energy conversion efficiencies range from 35.5 to 51.7%, corresponding to solar-to-fuel efficiencies of 11.7 to 17%, so even the least efficient route exceeds the 8% theoretical ceiling of algal photosynthesis. The serine variant of the reductive glycine pathway reaches an electrical energy conversion efficiency of when using H2 oxidation, within 1.9 points of the most efficient pathway, and is the only high-efficiency option that tolerates O2. This makes an EMP system that combines electron delivery by formate coupled with the serine variant of reductive glycine pathway highly attractive, as it presents few barriers to rapid, iterative engineering in the lab, and a high theoretical ceiling. Drawing both carbon and electrons from formate costs 6.2 points against H2 at a state-of-the-art whole-cell voltage (2.2 V). However, this small penalty is amplified three-fold by any rise in the CO2-to-formate cell voltage, and reaches 13.5 points at the highest whole-cell voltages reported for scaled-up CO2-to-formate electrolyzers, where formate-only operation falls to 11.2 electrical-to-fuel and 3.7% solar-to-fuel efficiency, below the ceiling of photosynthesis, against 24.7 and 8.1% for H2 (only just above algal photosynthesis). Our choice between a formate-only system and one coupled to H2 oxidation or EEU therefore depends on our belief about the trajectory of CO2 reduction technology. If whole-cell voltages continue to fall at the rate of the past decade, formate alone is the right target, and the simplicity of its workflow is bought at low cost. However, if that improvement plateaus, the electron delivery mechanism must be swappable, and a system should be designed from the outset so that it can be. At the US Department of Energy SunShot target of 2 cents per kilowatt hour, the electricity to make a US gallon of butanol costs $1.40 for a formate-only system at the state of the art, rising to $5.45 at the highest scaled-up electrolyzer voltage reported, against $1.23 and $2.47 for formate and H2 system.

synthetic biology↗
bioRxiv2026

Gene expression noise is reduced in communicating synthetic cell populations

A major goal in bottom-up synthetic biology is the construction of multicellular synthetic systems capable of coordinated and robust collective behaviours. However, robustness is often limited by noise and variability arising from increased molecular complexity. Whilst communication has been implemented in synthetic multi-cellular systems, the ability for communication to suppress cell free gene expression variability in populations of synthetic cells remain unexplored. To address this, we encapsulated the Lux and Las quorum sensing gene circuits in lipid vesicles under cell-free conditions to test the effect of communication on reducing cell-free gene expression variability across the population. Our results show that communication, limiting expression resources, and membrane surface effects can reduce gene expression variability. Resource limited Gillespie simulations for transcription and translation show that communication-mediated coupling reduces population-level expression noise under constrained and excess resource conditions. Together, our work provides simple strategies to reduce gene expression variability and thereby improve robustness in synthetic multicellular systems, an important criteria for the future applications of synthetic cells.

synthetic biology↗