Predicting retrosynthetic pathways using transformer-based models and a hyper-graph exploration strategy

We present an extension of our Molecular Transformer model combined with a hyper-graph exploration strategy for automatic retrosynthesis route planning without human intervention. The single-step retrosynthetic model sets a new state of the art for predicting reactants as well as reagents, solvents and catalysts for each retrosynthetic step. We introduce four metrics (coverage, class diversity, round-trip accuracy and Jensen–Shannon divergence) to evaluate the single-step retrosynthetic models, using the forward prediction and a reaction classification model always based on the transformer architecture. The hypergraph is constructed on the fly, and the nodes are filtered and further expanded based on a Bayesian-like probability. We critically assessed the end-to-end framework with several retrosynthesis examples from literature and academic exams. Overall, the frameworks have an excellent performance with few weaknesses related to the training data. The use of the introduced metrics opens up the possibility to optimize entire retrosynthetic frameworks by focusing on the performance of the single-step model only.

We present an extension of our Molecular Transformer model combined with a hyper-graph exploration strategy for automatic retrosynthesis route planning without human intervention.     

Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning

Computer aided synthesis planning (CASP) is part of a suite of artificial intelligence (AI) based tools that are able to propose synthesis routes to a wide range of compounds. However, at present they are too slow to be used to screen the synthetic feasibility of millions of generated or enumerated compounds before identification of potential bioactivity by virtual screening (VS) workflows. Herein we report a machine learning (ML) based method capable of classifying whether a synthetic route can be identified for a particular compound or not by the CASP tool AiZynthFinder. The resulting ML models return a retrosynthetic accessibility score (RAscore) of any molecule of interest, and computes at least 4500 times faster than retrosynthetic analysis performed by the underlying CASP tool. The RAscore should be useful for pre-screening millions of virtual molecules from enumerated databases or generative models for synthetic accessibility and produce higher quality databases for virtual screening of biological activity.

The retrosynthetic accessibility score (RAscore) is based on AI driven retrosynthetic planning, and is useful for rapid scoring of synthetic feasability and pre-screening of large datasets of virtual/generated molecules.     

Dynamic spatial and structural organization in artificial cells regulates signal processing by protein scaffolding

Structural and spatial organization are fundamental properties of biological systems that allow cells to regulate a wide range of biochemical processes. This organization is often transient and governed by external cues that initiate dynamic self-assembly processes. The construction of synthetic cell-like materials with similar properties requires the hierarchical and reversible organization of selected functional components on molecular scaffolds to dynamically regulate signaling pathways. The realization of such transient molecular programs in synthetic cells, however, remains underexplored due to the associated complexity of such hierarchical platforms. In this contribution, we effectuate dynamic spatial organization of effector protein subunits in a synthetic biomimetic compartment, a giant unilamellar vesicle (GUV), by associating in a reversible manner two fragments of a split luciferase to the membrane. This induces their structural dimerization, which consequently leads to the activation of enzymatic signaling. Importantly, such organization and activation are dynamic processes, and can be autonomously regulated – thus opening up avenues toward continuous spatiotemporal control over supramolecular organization and signaling in an artificial cell.

Engineered artificial cells respond to environmental cues through a pre-programmed enzymatic machinery that induces spatio-structural organization and activation of effector proteins at the lipid membrane.     

Bouncing off walls – widths of exit channels from shallow minima can dominate selectivity control

A selectivity model based on the widths of pathways to competing products, rather than barrier heights, is formulated for the butadiene + allyl cation reaction. This model was arrived at via analysis of stationary points, intrinsic reaction coordinates, potential energy surface shapes and direct dynamics trajectories, all determined using quantum chemical methods.

A selectivity model based on the widths of pathways to competing products, rather than barrier heights, is formulated for the butadiene + allyl cation reaction.     

Towards efficient discovery of green synthetic pathways with Monte Carlo tree search and reinforcement learning

Computer aided synthesis planning of synthetic pathways with green process conditions has become of increasing importance in organic chemistry, but the large search space inherent in synthesis planning and the difficulty in predicting reaction conditions make it a significant challenge. We introduce a new Monte Carlo Tree Search (MCTS) variant that promotes balance between exploration and exploitation across the synthesis space. Together with a value network trained from reinforcement learning and a solvent-prediction neural network, our algorithm is comparable to the best MCTS variant (PUCT, similar to Google''s Alpha Go) in finding valid synthesis pathways within a fixed searching time, and superior in identifying shorter routes with greener solvents under the same search conditions. In addition, with the same root compound visit count, our algorithm outperforms the PUCT MCTS by 16% in terms of determining successful routes. Overall the success rate is improved by 19.7% compared to the upper confidence bound applied to trees (UCT) MCTS method. Moreover, we improve 71.4% of the routes proposed by the PUCT MCTS variant in pathway length and choices of green solvents. The approach generally enables including Green Chemistry considerations in computer aided synthesis planning with potential applications in process development for fine chemicals or pharmaceuticals.

A new MCTS variant with a reinforcement learning value network and solvent prediction model proposes shorter synthesis routes with greener solvents.     

Systematic quantification of the dynamics of newly synthesized proteins unveiling their degradation pathways in human cells

Proteins are continuously synthesized during cell growth and proliferation. At the same time, excessive and misfolded proteins have to be degraded, otherwise they are a burden to cells. Protein degradation is essential to maintain proteostasis in cells, and dysfunction of protein degradation systems results in numerous diseases such as cancer and neurodegenerative diseases. Despite the importance of protein degradation, the degradation pathways of many proteins remain to be explored. Here, we comprehensively investigated the degradation of newly synthesized proteins in human cells by integrating metabolic labeling, click chemistry, and multiplexed proteomics, and systematic and quantitative analysis of newly synthesized proteins first revealed the degradation pathways of many proteins. Bioinformatic analysis demonstrates that proteins degraded through two major pathways have distinct properties and functions. Proteins degraded through the ubiquitin-proteasome pathway contain more disordered structures, whereas those through the autophagy-lysosome pathway have significantly higher hydrophobicity. Systematic and quantitative investigation of the dynamics of newly synthesized proteins provides unprecedented and valuable information about protein degradation, which leads to a better understanding of protein properties and cellular activities.

Systematic quantification of the dynamics of newly synthesized proteins first reveals the degradation pathways of many proteins in human cells, and proteins degraded through each of the two major pathways have distinct properties and functions.     

Chemical control of peptide material phase transitions

Progressive solute-rich polymer phase transitions provide pathways for achieving ordered supramolecular assemblies. Intrinsically disordered protein domains specifically regulate information in biological networks via conformational ordering. Here we consider a molecular tagging strategy to control ordering transitions in polymeric materials and provide a proof-of-principle minimal peptide phase network captured with a dynamic chemical network.

Substrate initiated assembly of a dynamic chemical network.     

Restoration of catalytic activity by the preservation of ligand structure: Cu-catalysed asymmetric conjugate addition with 1,1-diborylmethane

Reported herein is a novel reaction engineering protocol to enhance the efficiency of a transition metal-catalysed process by strategically preventing ligand degradation. Based on spectroscopic investigations, a decomposition pathway of a chiral phosphoramidite ligand during a Cu-catalysed reaction was identified. The involvement of the destructive process could be minimized under the modified reaction conditions that control the amount of nucleophilic alkoxide base, which is the origin of ligand decomposition. Overall, the strategy has been successfully applied to a new class of asymmetric conjugate addition reactions with bis[(pinacolato)boryl]methane, in which α,β-unsaturated enones are utilised as substrates.

A novel Cu-catalysed asymmetric conjugate addition reaction with bis[(pinacolato)boryl]methane using α,β-unsaturated enones as substrates has been developed on the basis of strategic preservation of the supporting ligand.     

DeepFrag: a deep convolutional neural network for fragment-based lead optimization

Machine learning has been increasingly applied to the field of computer-aided drug discovery in recent years, leading to notable advances in binding-affinity prediction, virtual screening, and QSAR. Surprisingly, it is less often applied to lead optimization, the process of identifying chemical fragments that might be added to a known ligand to improve its binding affinity. We here describe a deep convolutional neural network that predicts appropriate fragments given the structure of a receptor/ligand complex. In an independent benchmark of known ligands with missing (deleted) fragments, our DeepFrag model selected the known (correct) fragment from a set over 6500 about 58% of the time. Even when the known/correct fragment was not selected, the top fragment was often chemically similar and may well represent a valid substitution. We release our trained DeepFrag model and associated software under the terms of the Apache License, Version 2.0.

DeepFrag is a machine-learning model designed to assist with lead optimization. It recommends appropriate fragment additions given the 3D structures of a protein receptor and bound small-molecule ligand.     

C4-arylation and domino C4-arylation/3,2-carbonyl migration of indoles by tuning Pd catalytic modes: Pd(i)–Pd(ii) catalysis vs. Pd(ii) catalysis

Efficient C4-arylation and domino C4-arylation/3,2-carbonyl migration of indoles have been developed. The former route enables C4-arylation in a highly efficient and mild manner and the latter route provides an alternative straightforward protocol for synthesis of C2/C4 disubstituted indoles. The mechanism studies imply that the different reaction pathways were tuned by the distinct acid additives, which led to either the Pd(i)–Pd(ii) pathway or Pd(ii) catalysis.

C4-arylation via Pd(i)–Pd(ii) catalysis and domino C4-arylation/3,2-carbonyl migration of indoles via Pd(ii) catalysis tuning by acids have been developed.     

Exploring modular reengineering strategies to redesign the teicoplanin non-ribosomal peptide synthetase

Non-ribosomal peptide synthesis is an important biosynthesis pathway in secondary metabolism. In this study we have investigated modularisation and redesign strategies for the glycopeptide antibiotic teicoplanin. Using the relocation or exchange of domains within the NRPS modules, we have identified how to initiate peptide biosynthesis and explored the requirements for the functional reengineering of both the condensation/adenylation domain and epimerisation/condensation domain interfaces. We have also demonstrated strategies that ensure communication between isolated NRPS modules, leading to new peptide assembly pathways. This provides important insights into NRPS reengineering of glycopeptide antibiotic biosynthesis and has broad implications for the redesign of other NRPS systems.

Redesign of the non-ribosomal peptide synthetase (NRPS) from teicoplanin biosynthesis has been extensively investigated via domain exchange, interface reengineering and through engineering communication between isolated NRPS modules.     

Generation and application of Cu-bound alkyl nitrenes for the catalyst-controlled synthesis of cyclic β-amino acids

The advent of saturated N-heterocycles as valuable building blocks in medicinal chemistry has led to the development of new methods to construct such nitrogen-containing cyclic frameworks. Despite the apparent strategic clarity, intramolecular C–H aminations with metallonitrenes have only sporadically been explored in this direction because of the intractability of the requisite alkyl nitrenes. Here, we report copper-catalysed intramolecular amination using an alkyl nitrene generated from substituted isoxazolidin-5-ones upon N–O bond cleavage. The copper catalysis exclusively aminates aromatic C(sp²)–H bonds among other potentially reactive groups, offering a solution to the chemoselectivity problem that has been troublesome with rhodium catalysis. A combined experimental and computational study suggested that the active species in the current cyclic β-amino acid synthesis is a dicopper alkyl nitrene, which follows a cyclisation pathway distinct from the analogous alkyl metallonitrene.

Copper-catalysed conditions have been developed for the chemoselective synthesis of cyclic β-amino acids.     

Leveraging autocatalytic reactions for chemical domain image classification

Autocatalysis is fundamental to many biological processes, and kinetic models of autocatalytic reactions have mathematical forms similar to activation functions used in artificial neural networks. Inspired by these similarities, we use an autocatalytic reaction, the copper-catalyzed azide–alkyne cycloaddition, to perform digital image recognition tasks. Images are encoded in the concentration of a catalyst across an array of liquid samples, and the classification is performed with a sequence of automated fluid transfers. The outputs of the operations are monitored using UV-vis spectroscopy. The growing interest in molecular information storage suggests that methods for computing in chemistry will become increasingly important for querying and manipulating molecular memory.

Kinetic models of autocatalytic reactions have mathematical forms similar to activation functions used in artificial neural networks. Inspired by these similarities, we use a copper-catalyzed reaction to perform digital image recognition tasks.     

Cyclodepsipeptide alveolaride C: total synthesis and structural assignment

First stereoselective total synthesis of naturally occurring bioactive cyclodepsipeptide alveolaride C has been achieved using a convergent approach. This synthetic study enabled us to establish unambiguously the stereochemistry of three unassigned chiral centres embedded in the nonpeptidic segment as well as revised the stereochemistry of the proposed β-phenylalanine counterpart of the molecule. The key strategic features of this synthesis include Sharpless asymmetric dihydroxylation for installing the vicinal diol moiety, Julia–Kocienski olefination for constructing the aliphatic side chain, the Shiina protocol for intermolecular esterification, amide coupling and macrolactamization for the ring formation.

First total synthesis of natural cyclodepsipeptide alveolaride C has been accomplished with an unambiguous solution to its structural riddle.     

Hydrophosphination of boron–boron multiple bonds

Five compounds containing boron–boron multiple bonds are shown to undergo hydrophosphination reactions with diphenylphosphine in the absence of a catalyst. With diborenes, the products obtained are highly dependent on the substitution pattern at the boron atoms, with both 1,1- and 1,2-hydrophosphinations observed. With a symmetrical diboryne, 1,2-hydrophosphination yields a hydro(phosphino)diborene. The different mechanistic pathways for the hydrophosphination of diborenes are rationalised with the aid of density functional theory calculations.

Compounds containing boron–boron double and triple bonds are shown to undergo uncatalysed hydrophosphination reactions with diphenylphosphine.     

Snapshotting the transient conformations and tracing the multiple pathways of single peptide folding using a solid-state nanopore

A fundamental question relating to protein folding/unfolding is the time evolution of the folding of a protein into its precisely defined native structure. The proper identification of transition conformations is essential for accurately describing the dynamic protein folding/unfolding pathways. Owing to the rapid transitions and sub-nm conformation differences involved, the acquisition of the transient conformations and dynamics of proteins is difficult due to limited instrumental resolution. Using the electrochemical confinement effect of a solid-state nanopore, we were able to snapshot the transient conformations and trace the multiple transition pathways of a single peptide inside a nanopore. By combining the results with a Markov chain model, this new single-molecule technique is applied to clarify the transition pathways of the β-hairpin peptide, which shows nonequilibrium fluctuations among several blockage current stages. This method enables the high-throughput investigation of transition pathways experimentally to access previously obscure peptide dynamics, which is significant for understanding the folding/unfolding mechanisms and misfolding of peptides or proteins.

A solid-state nanopore based method is described for resolving protein-folding-related problems via snapshotting the folding intermediates and characterizing the kinetics of a single peptide.     

Dissipative self-assembly,competition and inhibition in a self-reproducing protocell model

The bottom-up synthesis of artificial, life-like systems promises to enable the study of emergent properties distinctive to life. Here, we report protocell systems generated from phase-separated building blocks. Vesicle protocells self-reproduce through a phase-transfer mechanism, catalysing their own formation. Dissipative self-assembly by the protocells is achieved when a hydrolysis step to destroy the surfactant is introduced. Competition between micelle and vesicle based replicators for a common feedstock shows that environmental conditions can control what species predominates: under basic conditions vesicles predominate, but in a neutral medium micelles are selected for via a mechanism which inhibits vesicle formation. Finally, the protocells enable orthogonal reactivity by catalysing in situ formation of an amphiphilic organocatalyst, which after incorporation into the vesicle bilayer enantioselectively forms a secondary product.

The bottom-up synthesis of a self-reproducing protocell model enables the study of emergent properties distinctive to life.     

Hydrogen and proton exchange at carbon. Imbalanced transition state and mechanism crossover

A recent remarkable study of the C–H oxidation of substituted fluorenyl-benzoates together with the transfer of a proton to an internal receiving group by means of electron transfer outer-sphere oxidants, in the noteworthy absence of hydrogen-bonding interactions, is taken as an example to uncover the existence of a mechanism crossover, making the reaction pass from a CPET pathway to a PTET pathway as the driving force of the global reaction decreases. This was also the occasion to stress that considerations based on “imbalanced” or “asynchronous” transition states cannot replace activation/driving force models based on the quantum mechanical treatment of both electrons and transferring protons.

Using the remarkable study of C–H oxidation of substituted fluorenyl-benzoates as an example, we have shown that a mechanism crossover takes place upon decreasing the driving force, from a CPET pathway to a PTET pathway.     

Construction of coacervate-in-coacervate multi-compartment protocells for spatial organization of enzymatic reactions

Coacervate microdroplets, formed via liquid–liquid phase separation, have been extensively explored as a compartment model for the construction of artificial cells or organelles. In this study, coacervate-in-coacervate multi-compartment protocells were constructed using four polyelectrolytes, in which carboxymethyl-dextran and diethylaminoethyl-dextran were deposited on the surface of as-prepared polydiallyldimethyl ammonium/deoxyribonucleic acid coacervate microdroplets through layer-by-layer assembly. The resulting multi-compartment protocells were composed from two immiscible coacervate phases with distinct physical and chemical properties. Molecule transport experiments indicated that small molecules could diffuse between two coacervate phases and that macromolecular enzymes could be retained. Furthermore, a competitive cascade enzymatic reaction of glucose oxidase/horseradish peroxidase–catalase was performed in the multi-compartment protocells. The different enzyme organization and productions of H₂O₂ led to a distinct polymerization of dopamine. The spatial organization of different enzymes in immiscible coacervate phases, the distinct reaction fluxes between coacervate phases, and the enzymatic cascade network led to distinguishable signal generation and product outputs. The development of this multi-compartment structure could pave the way toward the spatial organization of multi-enzyme cascades and provide new ideas for the design of organelle-containing artificial cells.

A coacervate-in-coacervate micro-architecture is constructed as a multi-compartment protocell model, in which a multi-enzyme cascade is spatially organized for competitive enzymatic reactions.