GitLab upgrade completed. Current version is 17.10.3. We now benefit from the enhancements of 17.9 and 17.10 releases. Among other improvements, it is now possible to set the automatic deletion of continuous integration pipelines. You can help us reduce storage usage by setting an expiry date for your pipelines.
Applications in biotechnology and bio-medical research make it desirable to design novel RNAs with very specific properties. Advanced design tasks require support by computational design but at the same time put high demands on the flexibility of computational tools and their expressivity to model the applications-specific requirements. To address such demands, we present the computational framework
Infrared that supports to develop advanced customized design tools, which generate designs with specific properties, often in a few lines of code.
Applications in biotechnology and bio-medical research call for effective strategies to design novel RNAs with very specific properties. Such advanced design tasks require support by computational design but at the same time put high demands on the flexibility of computational tools and their expressivity to model the applications-specific requirements. To address such demands, we present the computational framework \Infrared. It supports to develop advanced customized design tools, which generate designs with specific properties, often in a few lines of Python code.
This text guides the reader in tutorial-format through the development of a complex multi-target RNA design application. Thanks to \Infrared, it can be described as a step-by-step extension of an elementary design task: generating sequences that are compatible with a single RNA structure.
Finally, we point to further possibilities beyond the presented examples and discuss properties.
\TODO{Include some stochastic optimization of designs, such that we can directly target negative criteria?
Can we get some support from infrared (without already discussing the multi=target move set OR should we even include it)? This would allow us to replace the RNAdesign script in the Vienna RNA package...}
\section{Introduction}
\newcommand{\Def}[1]{{\bfseries #1}}
\TODO{some similarity to the Redprint bookchapter; add ideas about declarative application development}
Designing molecules with novel functionality or very specific desirable properties for applications in biological fundamental research, biotechnology, or even medicine, is a highly complex task that typically requires interdisciplinary efforts, combining biochemical experimentation and computational design. In several ways, designing RNAs can be even more attractive than designing proteins. On the one hand, functional RNA molecules save the detour of translation into proteins, and can therefore act more efficiently, e.g. as fast on/off-switches of gene activity. On the other hand, the design process itself can build on the well-understood combinatorics of RNA secondary structure and available computational models and algorithms.
Still, the supporting RNA design computationally is highly demanding. First of all, RNA design is an optimization problem with often complex objectives... multiple target structures ...
Moreover, RNA design is computationally complex even in simple problem variants. For example, one cannot efficiently design an RNA that preferentially folds into a single given target structure in the nearest-neighbor energy model, since this problem is NP-hard.
Here we present the framework \Infrared, which addresses the multiple demands of computational RNA design in several ways:
\begin{itemize}
\item To address the issues of computational complexity, it follows the classical decomposition of RNA design into two related sub-problems, often called positive and negative design. Positive RNA design aims at very specific properties (e.g. specific energy) for certain `target' RNA structures, while negative design additionally aims to avoid similar properties for all (exponentially many) other, `non-target' structures. While efficient algorithms for the latter problem can not exist,
the system automatically derives fixed-parameter tractable algorithms to solve the (sub-problem) positive design efficiently. In many cases, this can already solve negative design by searching through relatively few samples of good positive designs. To address harder negative design tasks, we demonstrate constraint generation as well as classic stochastic optimization techniques as supported by the system.
\item Real-world RNA design applications typically demand for targeting thermodynamic criteria referring to multiple target structures (e.g. on and off-states of riboswitches, binding pockets of aptamers and competing structures in specific energetic relations), potentially under side constraints like moderate GC-content or avoidance of specific sequence motifs. Thus, we design Infrared as a library that supports programmers to develop complex and potentially novel design strategies based on a declarative constraint framework. This allows application programmers to harness the power of fixed-parameter tractable sampling of designs in an easy to use system.
