Llamol: a dynamic multi-conditional generative transformer for de novo molecular design.

Generative models have demonstrated substantial promise in Natural Language Processing (NLP) and have found application in designing molecules, as seen in General Pretrained Transformer (GPT) models. In our efforts to develop such a tool for exploring the organic chemical space in search of potentially electro-active compounds, we present Llamol, a single novel generative transformer model based on the Llama 2 architecture, which was trained on a 12.5M superset of organic compounds drawn from diverse public sources. To allow for a maximum flexibility in usage and robustness in view of potentially incomplete data, we introduce Stochastic Context Learning (SCL) as a new training procedure. We demonstrate that the resulting model adeptly handles single- and multi-conditional organic molecule generation with up to four conditions, yet more are possible. The model generates valid molecular structures in SMILES notation while flexibly incorporating three numerical and/or one token sequence into the generative process, just as requested. The generated compounds are very satisfactory in all scenarios tested. In detail, we showcase the model's capability to utilize token sequences for conditioning, either individually or in combination with numerical properties, making Llamol a potent tool for de novo molecule design, easily expandable with new properties. SCIENTIFIC CONTRIBUTION: We developed a novel generative transformer model, Llamol, based on the Llama 2 architecture that was trained on a diverse set of 12.5 M organic compounds. It introduces Stochastic Context Learning (SCL) as a new training procedure, allowing for flexible and robust generation of valid organic molecules with up to multiple conditions that can be combined in various ways, making it a potent tool for de novo molecular design.

Insights into the Folding of Disulfide-Rich µ-Conotoxins.

The study of protein conformations using molecular dynamics (MD) simulations has been in place for decades. A major contribution to the structural stability and native conformation of a protein is made by the primary sequence and disulfide bonds formed during the folding process. Here, we investigated µ-conotoxins GIIIA, KIIIA, PIIIA, SIIIA, and SmIIIA as model peptides possessing three disulfide bonds. Their NMR structures were used for MD simulations in a novel approach studying the conformations between the folded and the unfolded states by systematically breaking the distinct disulfide bonds and monitoring the conformational stability of the peptides. As an outcome, the use of a combination of the existing knowledge and results from the simulations to classify the studied peptides within the extreme models of disulfide folding pathways, namely the bovine pancreatic trypsin inhibitor pathway and the hirudin pathway, is demonstrated. Recommendations for the design and synthesis of cysteine-rich peptides with a reduced number of disulfide bonds conclude the study.

The octet rule in chemical space: generating virtual molecules.

We present a generator of virtual molecules that selects valid chemistry on the basis of the octet rule. Also, we introduce a mesomer group key that allows a fast detection of duplicates in the generated structures. Compared to existing approaches, our model is simpler and faster, generates new chemistry and avoids invalid chemistry. Its versatility is illustrated by the correct generation of molecules containing third-row elements and a surprisingly adept handling of complex boron chemistry. Without any empirical parameters, our model is designed to be valid also in unexplored regions of chemical space. One first unexpected finding is the high prevalence of dipolar structures among generated molecules.

On the nature of interactions between ionic liquids and small amino-acid-based biomolecules.

During the last decade, ionic liquids (ILs) have revealed promising properties and applications in many research fields, including biotechnology and biological sciences. The focus of this contribution is to give a critical review of the phenomena observed and current knowledge of the interactions occurring on a molecular basis. As opposed to the huge advances made in understanding the properties of proteins in ILs, complementary investigations dealing with interactions between ILs and peptides or oligopeptides are underrepresented and are mostly only of phenomenological nature. However, the field has received more attention in the last few years. This Review features a meta-analysis of the available data and findings and should, therefore, provide a basis for a scientifically profound understanding of the nature and mechanisms of interactions between ILs and structured or nonstructured peptides. Fundamental aspects of the interactions between different peptides/oligopeptides and ILs are complemented by sections on the experimental (spectroscopy, structural biology) and theoretical (computational chemistry) possibilities to explain the phenomena reported so far in the literature. In effect, this should lead to the development of novel applications and support the understanding of IL-solute interactions in general.

Ligand-specific binding and activation of the human adenosine A(2B) receptor.

Adenosine A(2B) receptors, which play a role in inflammation and cancer, are of considerable interest as novel drug targets. To gain deeper insights into ligand binding and receptor activation, we exchanged amino acids predicted to be close to the binding pocket. The alanine mutants were stably expressed in CHO cells and characterized by radioligand binding and cAMP assays using three structural classes of ligands: xanthine (antagonist), adenosine, and aminopyridine derivatives (agonists). Asn282(7.45) and His280(7.43) were found to stabilize the binding site by intramolecular hydrogen bond formation as in the related A(2A) receptor subtype. Trp247(6.48), Val250(6.51), and particularly Ser279(7.42) were shown to be important for binding of nucleosidic agonists. Leu81(3.28), Asn186(5.42), and Val250(6.51) were discovered to be crucial for binding of the xanthine-derived antagonist PSB-603. Leu81(3.28), which is not conserved among adenosine receptor subtypes, may be important for the high selectivity of PSB-603. The N186(5.42)A mutant resulted in an increased potency for agonists. The interactions of the non-nucleosidic agonist BAY60-6583 were different from those of the nucleosides: while BAY60-6583 appeared not to interact with Ser279(7.42), its interactions with Trp247(6.48) and Val250(6.51) were significantly weaker compared to those of NECA. Moreover, our results discount the hypothesis of Trp247(6.48) serving as a "toogle switch" because BAY60-6583 was able to activate the corresponding mutant. This study reveals distinct interactions of structurally diverse ligands with the human A(2B) receptor and differences between closely related receptor subtypes (A(2B) and A(2A)). It will contribute to the understanding of G protein-coupled receptor function and advance A(2B) receptor ligand design.

A Glycam-Based Force Field for Simulations of Lipopolysaccharide Membranes: Parametrization and Validation.

Lipopolysaccharides (LPS) comprise the outermost layer of the Gram-negative bacteria cell envelope. Packed onto a lipid layer, the outer membrane displays remarkable physical-chemical differences compared to cell membranes. The carbohydrate-rich region confers a membrane asymmetry that underlies many biological processes such as endotoxicity, antibiotic resistance, and cell adhesion. Furthermore, unlike membrane proteins from other sources, integral outer-membrane proteins do not consist of transmembrane α helices; instead they consist of antiparallel ß-barrels, which highlights the importance of the LPS membrane as a medium. In this work, we present an extension of the GLYCAM06 force field that has been specifically developed for LPS membranes using our Wolf2Pack program. This new set of parameters for lipopolysaccharide molecules expands the GLYCAM06 repertoire of monosaccharides to include phosphorylated N- and O-acetylglucosamine, 3-deoxy-d-manno-oct-2-ulosonic acid, l-glycero-D-manno-heptose and its O-carbamoylated variant, and N-alanine-d-galactosamine. A total of 1 µs of molecular dynamics simulations of the rough LPS membrane of Pseudomonas aeruginosa PA01 is used to showcase the added parameter set. The equilibration of the LPS membrane is shown to be significantly slower compared to phospholipid membranes, on the order of 500 ns. It is further shown that water molecules penetrate the hydrocarbon region up to the terminal methyl groups, much deeper than commonly observed for phospholipid bilayers, and in agreement with neutron diffraction measurements. A comparison of simulated structural, dynamical, and electrostatic properties against corresponding experimentally available data shows that the present parameter set reproduces well the overall structure and the permeability of LPS membranes in the liquid-crystalline phase.

The four cysteine residues in the second extracellular loop of the human adenosine A2B receptor: role in ligand binding and receptor function.

The adenosine A(2B) receptor is of considerable interest as a new drug target for the treatment of asthma, inflammatory diseases, pain, and cancer. In the present study we investigated the role of the cysteine residues in the extracellular loop 2 (ECL2) of the receptor, which is particularly cysteine-rich, by a combination of mutagenesis, molecular modeling, chemical and pharmacological experiments. Pretreatment of CHO cells recombinantly expressing the human A(2B) receptor with dithiothreitol led to a 74-fold increase in the EC(50) value of the agonist NECA in cyclic AMP accumulation. In the C78(3.25)S and the C171(45.50)S mutant high-affinity binding of the A(2B) antagonist radioligand [(3)H]PSB-603 was abolished and agonists were virtually inactive in cAMP assays. This indicates that the C3.25-C45.50 disulfide bond, which is highly conserved in GPCRs, is also important for binding and function of A(2B) receptors. In contrast, the C166(45.45)S and the C167(45.46)S mutant as well as the C166(45.45)S-C167(45.46)S double mutant behaved like the wild-type receptor, while in the C154(45.33)S mutant significant, although more subtle effects on cAMP accumulation were observed - decrease (BAY60-6583) or increase (NECA) - depending on the structure of the investigated agonist. In contrast to the X-ray structure of the closely related A(2A) receptor, which showed four disulfide bonds, the present data indicate that in the A(2B) receptor only the C3.25-C45.50 disulfide bond is essential for ligand binding and receptor activation. Thus, the cysteine residues in the ECL2 of the A(2B) receptor not involved in stabilization of the receptor structure may have other functions.

Folding and unfolding characteristics of short beta strand peptides under different environmental conditions and starting configurations.

We analyze the effect of different environmental conditions, sequence lengths and starting configurations on the folding and unfolding pathways of small peptides exhibiting beta turns. We use chignolin and a sequence of peptide G as examples. A variety of different analysis tools allows us to characterize the changes in the folding pathways. It is observed that different harmonic modes dominate not only for different conditions but also for different starting points. The modes remain essentially very similar but their relative importance varies. A detailed analysis from diverse viewpoints including the influence of the particular amino acid sequence, conformational aspects as well as the associated motions yields a global picture that is consistent with experimental evidence and theoretical studies published elsewhere. Patterns of modes that remain stable over a range of temperatures might serve as an additional diagnostic to identify conformations that have reliably adopted a native fold. This could aid in reconstructing the folding process of a complete protein by identifying conformationally determined regions.

Homology modelling of the human adenosine A2B receptor based on X-ray structures of bovine rhodopsin, the beta2-adrenergic receptor and the human adenosine A2A receptor.

A three-dimensional model of the human adenosine A2B receptor was generated by means of homology modelling, using the crystal structures of bovine rhodopsin, the beta2-adrenergic receptor, and the human adenosine A2A receptor as templates. In order to compare the three resulting models, the binding modes of the adenosine A2B receptor antagonists theophylline, ZM241385, MRS1706, and PSB601 were investigated. The A2A-based model was much better able to stabilize the ligands in the binding site than the other models reflecting the high degree of similarity between A2A and A2B receptors: while the A2B receptor shares about 21% of the residues with rhodopsin, and 31% with the beta2-adrenergic receptor, it is 56% identical to the adenosine A2A receptor. The A2A-based model was used for further studies. The model included the transmembrane domains, the extracellular and the intracellular hydrophilic loops as well as the terminal domains. In order to validate the usefulness of this model, a docking analysis of several selective and nonselective agonists and antagonists was carried out including a study of binding affinities and selectivities of these ligands with respect to the adenosine A2A and A2B receptors. A common binding site is proposed for antagonists and agonists based on homology modelling combined with site-directed mutagenesis and a comparison between experimental and calculated affinity data. The new, validated A2B receptor model may serve as a basis for developing more potent and selective drugs.

WISDOM-II: screening against multiple targets implicated in malaria using computational grid infrastructures.

BACKGROUND: Despite continuous efforts of the international community to reduce the impact of malaria on developing countries, no significant progress has been made in the recent years and the discovery of new drugs is more than ever needed. Out of the many proteins involved in the metabolic activities of the Plasmodium parasite, some are promising targets to carry out rational drug discovery. MOTIVATION: Recent years have witnessed the emergence of grids, which are highly distributed computing infrastructures particularly well fitted for embarrassingly parallel computations like docking. In 2005, a first attempt at using grids for large-scale virtual screening focused on plasmepsins and ended up in the identification of previously unknown scaffolds, which were confirmed in vitro to be active plasmepsin inhibitors. Following this success, a second deployment took place in the fall of 2006 focussing on one well known target, dihydrofolate reductase (DHFR), and on a new promising one, glutathione-S-transferase. METHODS: In silico drug design, especially vHTS is a widely and well-accepted technology in lead identification and lead optimization. This approach, therefore builds, upon the progress made in computational chemistry to achieve more accurate in silico docking and in information technology to design and operate large scale grid infrastructures. RESULTS: On the computational side, a sustained infrastructure has been developed: docking at large scale, using different strategies in result analysis, storing of the results on the fly into MySQL databases and application of molecular dynamics refinement are MM-PBSA and MM-GBSA rescoring. The modeling results obtained are very promising. Based on the modeling results, In vitro results are underway for all the targets against which screening is performed. CONCLUSION: The current paper describes the rational drug discovery activity at large scale, especially molecular docking using FlexX software on computational grids in finding hits against three different targets (PfGST, PfDHFR, PvDHFR (wild type and mutant forms) implicated in malaria. Grid-enabled virtual screening approach is proposed to produce focus compound libraries for other biological targets relevant to fight the infectious diseases of the developing world.

Endogenous tetrahydroisoquinolines associated with Parkinson's disease mimic the feedback inhibition of tyrosine hydroxylase by catecholamines.

N-methyl-norsalsolinol and related tetrahydroisoquinolines accumulate in the nigrostriatal system of the human brain and are increased in the cerebrospinal fluid of patients with Parkinson's disease. We show here that 6,7-dihydroxylated tetrahydroisoquinolines such as N-methyl-norsalsolinol inhibit tyrosine hydroxylase, the key enzyme in dopamine synthesis, by imitating the mechanisms of catecholamine feedback regulation. Docked into a model of the enzyme's active site, 6,7-dihydroxylated tetrahydroisoquinolines were ligated directly to the iron in the catalytic center, occupying the same position as the catecholamine inhibitor dopamine. In this position, the ligands competed with the essential tetrahydropterin cofactor for access to the active site. Electron paramagnetic resonance spectroscopy revealed that, like dopamine, 6,7-dihydroxylated tetrahydroisoquinolines rapidly convert the catalytic iron to a ferric (inactive) state. Catecholamine binding increases the thermal stability of tyrosine hydroxylase and improves its resistance to proteolysis. We observed a similar effect after incubation with N-methyl-norsalsolinol or norsalsolinol. Following an initial rapid decline in tyrosine hydroxylation, the residual activity remained stable for 5 h at 37 degrees C. Phosphorylation by protein kinase A facilitates the release of bound catecholamines and is the most prominent mechanism of tyrosine hydroxylase reactivation. Protein kinase A also fully restored enzyme activity after incubation with N-methyl-norsalsolinol, demonstrating that tyrosine hydroxylase inhibition by 6,7-dihydroxylated tetrahydroisoquinolines mimics all essential aspects of catecholamine end-product regulation. Increased levels of N-methyl-norsalsolinol and related tetrahydroisoquinolines are therefore likely to accelerate dopamine depletion in Parkinson's disease.

Design of new plasmepsin inhibitors: a virtual high throughput screening approach on the EGEE grid.

Though different species of the genus Plasmodium may be responsible for malaria, the variant caused by P. falciparum is often very dangerous and even fatal if untreated. Hemoglobin degradation is one of the key metabolic processes for the survival of the Plasmodium parasite in its host. Plasmepsins, a family of aspartic proteases encoded by the Plasmodium genome, play a prominent role in host hemoglobin cleavage. In this paper we demonstrate the use of virtual screening, in particular molecular docking, employed at a very large scale to identify novel inhibitors for plasmepsins II and IV. A large grid infrastructure, the EGEE grid, was used to address the problem of large computation resources required for docking hundreds of thousands of chemical compounds on different plasmepsin targets of P. falciparum. A large compound library of about 1 million chemical compounds was docked on 5 different targets of plasmepsins using two different docking software, namely FlexX and AutoDock. Several strategies were employed to analyze the results of this virtual screening approach including docking scores, ideal binding modes, and interactions to key residues of the protein. Three different classes of structures with thiourea, diphenylurea, and guanidino scaffolds were identified to be promising hits. While the identification of diphenylurea compounds is in accordance with the literature and thus provides a sort of "positive control", the identification of novel compounds with a guanidino scaffold proves that high throughput docking can be effectively used to identify novel potential inhibitors of P. falciparum plasmepsins. Thus, with the work presented here, we do not only demonstrate the relevance of computational grids in drug discovery but also identify several promising small molecules which have the potential to serve as candidate inhibitors for P. falciparum plasmepsins. With the use of the EGEE grid infrastructure for the virtual screening campaign against the malaria causing parasite P. falciparum we have demonstrated that resource sharing on an eScience infrastructure such as EGEE provides a new model for doing collaborative research to fight diseases of the poor.

Grid enabled high throughput virtual screening against four different targets implicated in malaria.

After having deployed a first data challenge on malaria and a second one on avian flu, respectively in summer 2005 and spring 2006, we are demonstrating here again how efficiently the computational grids can be used to produce massive docking data at a high-throughput. During more than 2 months and a half, we have achieved at least 140 million dockings, representing an average throughput of almost 80,000 dockings per hour. This was made possible by the availability of thousands of CPUs through different infrastructures worldwide. Through the acquired experience, the WISDOM production environment is evolving to enable an easy and fault-tolerant deployment of biological tools; in this case it is the FlexX commercial docking software which is used to dock the whole ZINC database against 4 different targets.

Demonstration of in silico docking at a large scale on grid infrastructure.

WISDOM stands for World-wide In Silico Docking On Malaria. First step toward enabling the in silico drug discovery pipeline on a grid infrastructure, this CPU consuming application generating large data flows was deployed successfully on EGEE, the largest grid infrastructure in the world, during the summer 2005. 46 million docking scores were computed in 6 weeks. The proposed demonstration presents the submission of in silico docking jobs at a large scale on the grid. The demonstration will use the new middleware stack gLite developed within the EGEE project.

Adenosine receptor agonists: from basic medicinal chemistry to clinical development.

Adenosine is a physiological nucleoside which acts as an autocoid and activates G protein-coupled membrane receptors, designated A(1), A(2A), A(2B) and A(3). Adenosine plays an important role in many (patho)physiological conditions in the CNS as well as in peripheral organs and tissues. Adenosine receptors are present on virtually every cell. However, receptor subtype distribution and densities vary greatly. Adenosine itself is used as a therapeutic agent for the treatment of supraventricular paroxysmal tachycardia and arrhythmias and as a vasodilatatory agent in cardiac imaging. During the past 20 years, a number of selective agonists for A(1), A(2A) and A(3) adenosine receptors have been developed, all of them structurally derived from adenosine. Several such compounds are currently undergoing clinical trials for the treatment of cardiovascular diseases (A(1)and A(2A)), pain (A(1)), wound healing (A(2A)), diabetic foot ulcers (A(2A)), colorectal cancer (A(3)) and rheumatoid arthritis (A(3)). Clinical evaluation of some A(1) and A(2A) adenosine receptor agonists has been discontinued. Major problems include side effects due to the wide distribution of adenosine receptors; low brain penetration, which is important for the targeting of CNS diseases; short half-lifes of compounds; or a lack of effects, in some cases perhaps due to receptor desensitisation or to low receptor density in the targeted tissue. Partial agonists, inhibitors of adenosine metabolism (adenosine kinase and deaminase inhibitors) or allosteric activators of adenosine receptors may be advantageous for certain indications, as they may exhibit fewer side effects.

Modeled ligand-protein complexes elucidate the origin of substrate specificity and provide insight into catalytic mechanisms of phenylalanine hydroxylase and tyrosine hydroxylase.

NMR spectroscopy and X-ray crystallography have provided important insight into structural features of phenylalanine hydroxylase (PAH) and tyrosine hydroxylase (TH). Nevertheless, significant problems such as the substrate specificity of PAH and the different susceptibility of TH to feedback inhibition by l-3,4-dihydroxyphenylalanine (l-DOPA) compared with dopamine (DA) remain unresolved. Based on the crystal structures 5pah for PAH and 2toh for TH (Protein Data Bank), we have used molecular docking to model the binding of 6(R)-l-erythro-5,6,7,8-tetrahydrobiopterin (BH4) and the substrates phenylalanine and tyrosine to the catalytic domains of PAH and TH. The amino acid substrates were placed in positions common to both enzymes. The productive position of tyrosine in TH.BH4 was stabilized by a hydrogen bond with BH4. Despite favorable energy scores, tyrosine in a position trans to PAH residue His290 or TH residue His336 interferes with the access of the essential cofactor dioxygen to the catalytic center, thereby blocking the enzymatic reaction. DA and l-DOPA were directly coordinated to the active site iron via the hydroxyl residues of their catechol groups. Two alternative conformations, rotated 180 degrees around an imaginary iron-catecholamine axis, were found for DA and l-DOPA in PAH and for DA in TH. Electrostatic forces play a key role in hindering the bidentate binding of the immediate reaction product l-DOPA to TH, thereby saving the enzyme from direct feedback inhibition.