The openOCHEM consensus model is the best-performing open-source predictive model in the First EUOS/SLAS joint compound solubility challenge.

The EUOS/SLAS challenge aimed to facilitate the development of reliable algorithms to predict the aqueous solubility of small molecules using experimental data from 100 K compounds. In total, hundred teams took part in the challenge to predict low, medium and highly soluble compounds as measured by the nephelometry assay. This article describes the winning model, which was developed using the publicly available Online CHEmical database and Modeling environment (OCHEM) available on the website https://ochem.eu/article/27. We describe in detail the assumptions and steps used to select methods, descriptors and strategy which contributed to the winning solution. In particular we show that consensus based on 28 models calculated using descriptor-based and representation learning methods allowed us to obtain the best score, which was higher than those based on individual approaches or consensus models developed using each individual approach. A combination of diverse models allowed us to decrease both bias and variance of individual models and to calculate the highest score. The model based on Transformer CNN contributed the best individual score thus highlighting the power of Natural Language Processing (NLP) methods. The inclusion of information about aleatoric uncertainty would be important to better understand and use the challenge data by the contestants.

State-of-the-art augmented NLP transformer models for direct and single-step retrosynthesis.

We investigated the effect of different training scenarios on predicting the (retro)synthesis of chemical compounds using text-like representation of chemical reactions (SMILES) and Natural Language Processing (NLP) neural network Transformer architecture. We showed that data augmentation, which is a powerful method used in image processing, eliminated the effect of data memorization by neural networks and improved their performance for prediction of new sequences. This effect was observed when augmentation was used simultaneously for input and the target data simultaneously. The top-5 accuracy was 84.8% for the prediction of the largest fragment (thus identifying principal transformation for classical retro-synthesis) for the USPTO-50k test dataset, and was achieved by a combination of SMILES augmentation and a beam search algorithm. The same approach provided significantly better results for the prediction of direct reactions from the single-step USPTO-MIT test set. Our model achieved 90.6% top-1 and 96.1% top-5 accuracy for its challenging mixed set and 97% top-5 accuracy for the USPTO-MIT separated set. It also significantly improved results for USPTO-full set single-step retrosynthesis for both top-1 and top-10 accuracies. The appearance frequency of the most abundantly generated SMILES was well correlated with the prediction outcome and can be used as a measure of the quality of reaction prediction.

GEN: highly efficient SMILES explorer using autodidactic generative examination networks.

Recurrent neural networks have been widely used to generate millions of de novo molecules in defined chemical spaces. Reported deep generative models are exclusively based on LSTM and/or GRU units and frequently trained using canonical SMILES. In this study, we introduce Generative Examination Networks (GEN) as a new approach to train deep generative networks for SMILES generation. In our GENs, we have used an architecture based on multiple concatenated bidirectional RNN units to enhance the validity of generated SMILES. GENs autonomously learn the target space in a few epochs and are stopped early using an independent online examination mechanism, measuring the quality of the generated set. Herein we have used online statistical quality control (SQC) on the percentage of valid molecular SMILES as examination measure to select the earliest available stable model weights. Very high levels of valid SMILES (95-98%) can be generated using multiple parallel encoding layers in combination with SMILES augmentation using unrestricted SMILES randomization. Our trained models combine an excellent novelty rate (85-90%) while generating SMILES with strong conservation of the property space (95-99%). In GENs, both the generative network and the examination mechanism are open to other architectures and quality criteria.

Transformer-CNN: Swiss knife for QSAR modeling and interpretation.

We present SMILES-embeddings derived from the internal encoder state of a Transformer [1] model trained to canonize SMILES as a Seq2Seq problem. Using a CharNN [2] architecture upon the embeddings results in higher quality interpretable QSAR/QSPR models on diverse benchmark datasets including regression and classification tasks. The proposed Transformer-CNN method uses SMILES augmentation for training and inference, and thus the prognosis is based on an internal consensus. That both the augmentation and transfer learning are based on embeddings allows the method to provide good results for small datasets. We discuss the reasons for such effectiveness and draft future directions for the development of the method. The source code and the embeddings needed to train a QSAR model are available on https://github.com/bigchem/transformer-cnn. The repository also has a standalone program for QSAR prognosis which calculates individual atoms contributions, thus interpreting the model's result. OCHEM [3] environment (https://ochem.eu) hosts the on-line implementation of the method proposed.

Reversible formation of aminals: a new strategy to control the release of bioactive volatiles from dynamic mixtures.

Dynamic mixtures generated by reversible aminal formation of fragrance aldehydes with N,N-dibenzyl alkyldiamines in aqueous systems were found to be suitable delivery systems for the controlled release of bioactive volatiles.

Evolving the substrate specificity of O6-alkylguanine-DNA alkyltransferase through loop insertion for applications in molecular imaging.

We introduce a strategy for evolving protein substrate specificity by the insertion of random amino acid loops into the protein backbone. Application of this strategy to human O6-alkylguanine-DNA alkyltransferase (AGT) led to the isolation of mutants that react with the non-natural substrate O6-propargylguanine. Libraries generated by conventional random or targeted saturation mutagenesis, by contrast, did not yield any mutants with activity towards this new substrate. The strategy of loop insertion to alter enzyme specificity should be general and applicable to other classes of proteins. An important application of the isolated AGT mutant is in molecular imaging, where the mutant and parental AGTs are used to label two different AGT fusion proteins with different fluorophores in the same living cell or in vitro . This allowed the establishment of fluorescence-based assays to detect protein-protein interactions and measure enzymatic activities.

Alpha-1-C-octyl-1-deoxynojirimycin as a pharmacological chaperone for Gaucher disease.

The most common lysosomal storage disorder, Gaucher disease, is caused by inefficient folding and trafficking of certain variants of lysosomal beta-glucosidase (beta-Glu, also known as beta-glucocerebrosidase). Recently, Sawker et al. reported that the addition of subinhibitory concentrations (10 microM) of the pharmacological chaperone N-nonyl-1-deoxynojirimycin (NN-DNJ) (10) to Gaucher patient-derived cells leads to a 2-fold increase in activity of mutant (N370S) enzyme [Proc. Natl. Acad. Sci. U.S.A.2002, 99, 15428]. However, we found that the addition of NN-DNJ at 10 microM lowered the lysosomal alpha-glucosidase (alpha-Glu) activity by 50% throughout the assay period in spite of the excellent chaperoning activity in N370S fibroblasts. Hence, we prepared a series of DNJ derivatives with an alkyl chain at the C-1alpha position and evaluated their in vitro inhibitory activity and potential as pharmacological chaperones for Gaucher cell lines. Among them, alpha-1-C-octyl-DNJ (CO-DNJ) (15) showed 460-fold stronger in vitro inhibitory activity than DNJ toward beta-Glu, while NN-DNJ enhanced in vitro inhibitory activity by 360-fold. Treatment with CO-DNJ (20 microM) for 4 days maximally increased intracellular beta-Glu activity by 1.7-fold in Gaucher N370 cell line (GM0037) and by 2.0-fold in another N370 cell line (GM00852). The addition of 20 microM CO-DNJ to the N370S (GM00372) culture medium for 10 days led to 1.9-fold increase in the beta-Glu activity without affecting the intracellular alpha-Glu activity for 10 days. Only CO-DNJ showed a weak beta-Glu chaperoning activity in the L444P type 2 variant, with 1.2-fold increase at 5-20 microM, and furthermore maximally increased the alpha-Glu activity by 1.3-fold at 20 microM. These experimental results suggest that CO-DNJ is a significant pharmacological chaperone for N370S Gaucher variants, minimizing the potential for undesirable side effects such as alpha-Glu inhibition.

Protein function microarrays based on self-immobilizing and self-labeling fusion proteins.

Protein microarrays are an attractive approach for the high-throughput analysis of protein function, but their impact on proteomics has been limited by the technical difficulties associated with their generation. Here we demonstrate that fusion proteins of O6-alkylguanine-DNA alkyltransferase (AGT) can be used for the simple and reliable generation of protein microarrays for the analysis of protein function. Important features of the approach are the selectivity of the covalent immobilization; this allows for direct immobilization of proteins out of cell extracts, and the option both to label and to immobilize AGT fusion proteins, which allows for direct screening for protein-protein interactions between different AGT fusion proteins. In addition to the identification of protein-protein interactions, AGT-based protein microarrays can be used for the characterization of small molecule-protein interactions or post-translational modifications. The potential of the approach was demonstrated by investigating the post-translational modification of acyl carrier protein (ACP) from E. coli by different phosphopantetheine transferases (PPTases), yielding insights into the role of selected ACP amino acids in the ACP-PPTase interaction.

Adding value to fusion proteins through covalent labelling.

Recombinant proteins are often expressed with an additional polypeptide the role of which is to aid purification, detection or functional studies. Recently, the role of these tags has been extended to mediate the labelling of the protein of interest with chemically diverse compounds. These approaches are of importance for protein science as they allow proteins to be equipped with properties that cannot be genetically encoded and also enable the use of a single fusion protein for several different applications. Several new approaches have been developed for the covalent labelling of fusion proteins both in living cells and in vitro.

Alpha-1-C-alkyl-1-deoxynojirimycin derivatives as potent and selective inhibitors of intestinal isomaltase: remarkable effect of the alkyl chain length on glycosidase inhibitory profile.

A series of alpha- and beta-1-C-alkyl-1-deoxynojirimycin derivatives was prepared and evaluated as glycosidase inhibitors. Biological assays showed a marked dependence of the selectivity and potency of the inhibitors upon the position of the alkyl chain (alpha-1-C-, beta-1-C- or N-alkyl derivatives). In addition, the efficiency of alpha-1-C-alkyl-1-deoxynojirimycin derivatives as intestinal isomaltase inhibitors increases with the length of the alkyl chain. The strongest inhibition was found for alpha-1-C -nonyl-1-deoxynojirimycin with an IC50=3.5 nM (25x more potent inhibitor than the shorter chain homologue carrying a C8 chain). These results demonstrate that subtle changes in the aglycon fragment may result in remarkable enzyme specificity.

General access to iminosugar C-glycoside building blocks by means of cross-metathesis: a gateway to glycoconjugate mimetics.

[reaction: see text] Cross-metathesis reactions of alpha-1-C-allyl-1-deoxynojirimycin derivatives 7a,b and various functionalized alkenes mediated by Grubbs's catalyst 3 are reported. The reactions showed reasonable to very good yields and excellent E/Z selectivity. This methodology allows the efficient and convergent synthesis of iminosugar C-glycosides with a great degree of structural diversity in the aglycone, opening the way to a variety of new glycoconjugate mimetics of biological interest.

A general strategy for the practical synthesis of nojirimycin C-glycosides and analogues. Extension to the first reported example of an iminosugar 1-phosphonate.

An efficient and versatile strategy for the synthesis of nojirimycin C-glycosides and related compounds with full stereocontrol is reported. The key steps of the process are the addition of organometallic reagents onto an L-sorbose-derived imine (13) followed by an internal reductive amination. The addition step, which controls the alpha- vs beta-configuration at the pseudoanomeric center in the final product, is highly diastereoselective (re-face addition), and the stereoselectivity can be effectively inverted by adding an external monodentate Lewis acid (si-face addition). The complete synthesis could be achieved in 10 steps only from commercially available 2,3;4,6-di-O-isopropylidene-alpha-L-sorbofuranose and provided alpha- or beta-1-C-substituted 1-deoxynojirimycin derivatives in 27-52% overall yield. The strategy was successfully extended to the first example of an iminosugar 1-phosphonate. The methodology provides access to a wide range of biologically relevant glycoconjugate mimetics in which the glycosidic function is replaced by an imino-C-glycosidic linkage.