DetSpace: a web server for engineering detectable pathways for bio-based chemical production

Abstract Tackling climate change challenges requires replacing current chemical industrial processes through the rational and sustainable use of biodiversity resources. To that end, production routes to key bio-based chemicals for the bioeconomy have been identified. However, their production still remains inefficient in terms of titers, rates, and yields; because of the hurdles found when scaling up. In order to make production more efficient, strategies like automated screening and dynamic pathway regulation through biosensors have been applied as part of strain optimization. However, to date, no systematic way exists to design a genetic circuit that is responsive to concentrations of a given target compound. Here, the DetSpace web server provides a set of integrated tools that allows a user to select and design a biological circuit that performs the sensing of a molecule of interest by its enzymatic conversion to a detectable molecule through a transcription factor. In that way, the DetSpace web server allows synthetic biologists to easily design biosensing routes for the dynamic regulation of metabolic pathways in applications ranging from genetic circuits design, screening, production, and bioremediation of bio-based chemicals, to diagnostics and drug delivery.


Supplementary Note Monte Carlo tree search algorithm:
As mentioned in the article, a Monte Carlo tree search algorithm has been used to obtain the detectable routes.In this section, we are going to detail how it works.
Input: First, we start from the producible and detectable pair, plus the reaction rules of RetroRules.These rules allow us to generate new metabolites until we obtain the complete route.The Python module RDKit, which specializes in cheminformatics, has been used for this purpose.

Short description:
The process to generate the new metabolites consists of launching our detectable against all the rules and writing down the new compounds obtained.Here is where the Monte Carlo algorithm comes into play, as this is an exponential expansion, so it is necessary to limit the search.Each new generation of metabolites adds more paths to the search tree and the aim of the algorithm is to find the most promising paths.It consists of four steps: selection, expansion, simulation, and backpropagation.

Steps:
1. Selection: Starting from the root node, we navigate through the tree until we reach the leaf nodes that have not yet been simulated.The nodes with the highest assigned score are chosen for this traversal, so that we always take the most promising path.2. Expansion: Once the node is selected, new compounds are generated using the rules.It is possible that these new metabolites are already in the tree, when this happens, they are simply omitted.The rest are added as child nodes of our selected compound.In case the selected node has a score of 1, we have already reached our target compound, so the algorithm would end here.3. Simulation: For all new nodes a score is calculated.This score consists of the Tanimoto similarity with the target metabolite.4. Backpropagation: Using the score of the new nodes, the information of the nodes on the path between the root and these new nodes is updated.To prevent the algorithm from delving too deeply into some branches and ignoring other promising ones, a correction factor has been added to the score.Nodes that have not been chosen for a long period of iterations gradually increase their score, so that at some point the algorithm will change branches and explore the others.5. Output: When the search has reached our target compound the algorithm returns the resulting tree and a file containing info from all the reactions taking part on it.
The computations were performed on the HPC cluster Garnatxa at Institute for Integrative Systems Biology (I2SysBio), I2SysBio is a mixed research center formed by University of Valencia (UV) and Spanish National Research Council (CSIC), and at the HPC cluster Rigel of the Universitat Politècnica de València.

Figure S1 :
Figure S1: According to DetSpace, eriodictyol can be downstream transformed into quercetin with associated allosteric transcription factor by a two-step transformation involving taxifolin; which can be detected by the transcription factor LmrA from Bacillus subtilis.This process can be detected by the transcription factor QdoR from Bacillus subtilis, belonging both to the repressor TetR family.

Figure S2 :
Figure S2: According to DetSpace, eriodictyol can be downstream transformed into catechin, which e.g. can be detected by the transcription factor LmrA from Bacillus subtilis, by a three-step transformation involving taxifolin and leucocyanidin.