Drug-Target Interactions Prediction at Scale: The Komet Algorithm with the LCIdb Dataset.

Drug-target interactions (DTIs) prediction algorithms are used at various stages of the drug discovery process. In this context, specific problems such as deorphanization of a new therapeutic target or target identification of a drug candidate arising from phenotypic screens require large-scale predictions across the protein and molecule spaces. DTI prediction heavily relies on supervised learning algorithms that use known DTIs to learn associations between molecule and protein features, allowing for the prediction of new interactions based on learned patterns. The algorithms must be broadly applicable to enable reliable predictions, even in regions of the protein or molecule spaces where data may be scarce. In this paper, we address two key challenges to fulfill these goals: building large, high-quality training datasets and designing prediction methods that can scale, in order to be trained on such large datasets. First, we introduce LCIdb, a curated, large-sized dataset of DTIs, offering extensive coverage of both the molecule and druggable protein spaces. Notably, LCIdb contains a much higher number of molecules than publicly available benchmarks, expanding coverage of the molecule space. Second, we propose Komet (Kronecker Optimized METhod), a DTI prediction pipeline designed for scalability without compromising performance. Komet leverages a three-step framework, incorporating efficient computation choices tailored for large datasets and involving the Nyström approximation. Specifically, Komet employs a Kronecker interaction module for (molecule, protein) pairs, which efficiently captures determinants in DTIs, and whose structure allows for reduced computational complexity and quasi-Newton optimization, ensuring that the model can handle large training sets, without compromising on performance. Our method is implemented in open-source software, leveraging GPU parallel computation for efficiency. We demonstrate the interest of our pipeline on various datasets, showing that Komet displays superior scalability and prediction performance compared to state-of-the-art deep learning approaches. Additionally, we illustrate the generalization properties of Komet by showing its performance on an external dataset, and on the publicly available LH benchmark designed for scaffold hopping problems. Komet is available open source at https://komet.readthedocs.io and all datasets, including LCIdb, can be found at https://zenodo.org/records/10731712.

Exploring isofunctional molecules: Design of a benchmark and evaluation of prediction performance.

Identification of novel chemotypes with biological activity similar to a known active molecule is an important challenge in drug discovery called 'scaffold hopping'. Small-, medium-, and large-step scaffold hopping efforts may lead to increasing degrees of chemical structure novelty with respect to the parent compound. In the present paper, we focus on the problem of large-step scaffold hopping. We assembled a high quality and well characterized dataset of scaffold hopping examples comprising pairs of active molecules and including a variety of protein targets. This dataset was used to build a benchmark corresponding to the setting of real-life applications: one active molecule is known, and the second active is searched among a set of decoys chosen in a way to avoid statistical bias. This allowed us to evaluate the performance of computational methods for solving large-step scaffold hopping problems. In particular, we assessed how difficult these problems are, particularly for classical 2D and 3D ligand-based methods. We also showed that a machine-learning chemogenomic algorithm outperforms classical methods and we provided some useful hints for future improvements.

On the Frustration to Predict Binding Affinities from Protein-Ligand Structures with Deep Neural Networks.

Accurate prediction of binding affinities from protein-ligand atomic coordinates remains a major challenge in early stages of drug discovery. Using modular message passing graph neural networks describing both the ligand and the protein in their free and bound states, we unambiguously evidence that an explicit description of protein-ligand noncovalent interactions does not provide any advantage with respect to ligand or protein descriptors. Simple models, inferring binding affinities of test samples from that of the closest ligands or proteins in the training set, already exhibit good performances, suggesting that memorization largely dominates true learning in the deep neural networks. The current study suggests considering only noncovalent interactions while omitting their protein and ligand atomic environments. Removing all hidden biases probably requires much denser protein-ligand training matrices and a coordinated effort of the drug design community to solve the necessary protein-ligand structures.

New insights into structure and function of bis-phosphinic acid derivatives and implications for CFTR modulation.

C407 is a compound that corrects the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) protein carrying the p.Phe508del (F508del) mutation. We investigated the corrector effect of c407 and its derivatives on F508del-CFTR protein. Molecular docking and dynamics simulations combined with site-directed mutagenesis suggested that c407 stabilizes the F508del-Nucleotide Binding Domain 1 (NBD1) during the co-translational folding process by occupying the position of the p.Phe1068 side chain located at the fourth intracellular loop (ICL4). After CFTR domains assembly, c407 occupies the position of the missing p.Phe508 side chain. C407 alone or in combination with the F508del-CFTR corrector VX-809, increased CFTR activity in cell lines but not in primary respiratory cells carrying the F508del mutation. A structure-based approach resulted in the synthesis of an extended c407 analog G1, designed to improve the interaction with ICL4. G1 significantly increased CFTR activity and response to VX-809 in primary nasal cells of F508del homozygous patients. Our data demonstrate that in-silico optimized c407 derivative G1 acts by a mechanism different from the reference VX-809 corrector and provide insights into its possible molecular mode of action. These results pave the way for novel strategies aiming to optimize the flawed ICL4-NBD1 interface.

Functional and Pharmacological Characterization of the Rare CFTR Mutation W361R.

Understanding the functional consequence of rare cystic fibrosis (CF) mutations is mandatory for the adoption of precision therapeutic approaches for CF. Here we studied the effect of the very rare CF mutation, W361R, on CFTR processing and function. We applied western blot, patch clamp and pharmacological modulators of CFTR to study the maturation and ion transport properties of pEGFP-WT and mutant CFTR constructs, W361R, F508del and L69H-CFTR, expressed in HEK293 cells. Structural analyses were also performed to study the molecular environment of the W361 residue. Western blot showed that W361R-CFTR was not efficiently processed to a mature band C, similar to F508del CFTR, but unlike F508del CFTR, it did exhibit significant transport activity at the cell surface in response to cAMP agonists. Importantly, W361R-CFTR also responded well to CFTR modulators: its maturation defect was efficiently corrected by VX-809 treatment and its channel activity further potentiated by VX-770. Based on these results, we postulate that W361R is a novel class-2 CF mutation that causes abnormal protein maturation which can be corrected by VX-809, and additionally potentiated by VX-770, two FDA-approved small molecules. At the structural level, W361 is located within a class-2 CF mutation hotspot that includes other mutations that induce variable disease severity. Analysis of the 3D structure of CFTR within a lipid environment indicated that W361, together with other mutations located in this hotspot, is at the edge of a groove which stably accommodates lipid acyl chains. We suggest this lipid environment impacts CFTR folding, maturation and response to CFTR modulators.

Targeting different binding sites in the CFTR structures allows to synergistically potentiate channel activity.

Recent evidence shows that combination of correctors and potentiators, such as the drug ivacaftor (VX-770), can significantly restore the functional expression of mutated Cystic Fibrosis Transmembrane conductance Regulator (CFTR), an anion channel which is mutated in cystic fibrosis (CF). The success of these combinatorial therapies highlights the necessity of identifying a broad panel of specific binding mode modulators, occupying several distinct binding sites at structural level. Here, we identified two small molecules, SBC040 and SBC219, which are two efficient cAMP-independent potentiators, acting at low concentration of forskolin with EC50 close to 1 µM and in a synergic way with the drug VX-770 on several CFTR mutants of classes II and III. Molecular dynamics simulations suggested potential SBC binding sites at the vicinity of ATP-binding sites, distinct from those currently proposed for VX-770, outlining SBC molecules as members of a new family of potentiators.

Unsolved severe chronic rhinosinusitis elucidated by extensive CFTR genotyping.

Severe chronic rhinosinusitis in children should alert clinicians and extensive CFTR genotyping should be performed. We propose that thorough clinical and functional assessment in severe chronic rhinosinusitis is valuable to discover rare mutations which could be treated by CFTR correctors to postpone pulmonary infection.

Combining theoretical and experimental data to decipher CFTR 3D structures and functions.

Cryo-electron microscopy (cryo-EM) has recently provided invaluable experimental data about the full-length cystic fibrosis transmembrane conductance regulator (CFTR) 3D structure. However, this experimental information deals with inactive states of the channel, either in an apo, quiescent conformation, in which nucleotide-binding domains (NBDs) are widely separated or in an ATP-bound, yet closed conformation. Here, we show that 3D structure models of the open and closed forms of the channel, now further supported by metadynamics simulations and by comparison with the cryo-EM data, could be used to gain some insights into critical features of the conformational transition toward active CFTR forms. These critical elements lie within membrane-spanning domains but also within NBD1 and the N-terminal extension, in which conformational plasticity is predicted to occur to help the interaction with filamin, one of the CFTR cellular partners.

Cis variants identified in F508del complex alleles modulate CFTR channel rescue by small molecules.

Molecules correcting the trafficking (correctors) and gating defects (potentiators) of the cystic fibrosis causing mutation c.1521_1523delCTT (p.Phe508del) begin to be a useful treatment for CF patients bearing p.Phe508del. This mutation has been identified in different genetic contexts, alone or in combination with variants in cis. Until now, 21 exonic variants in cis of p.Phe508del have been identified, albeit at a low frequency. The aim of this study was to evaluate their impact on the efficacy of CFTR-directed corrector/potentiator therapy (Orkambi). The analysis by minigene showed that two out of 15 cis variants tested increased exon skipping (c.609C > T and c.2770G > A). Four cis variants were studied functionally in the absence of p.Phe508del, one of which was found to be deleterious for protein maturation c.1399C > T (p.Leu467Phe). In the presence of p.Phe508del, this variant was the only to prevent the response to Orkambi treatment. This study showed that some patients carrying p.Phe508del complex alleles are predicted to poorly respond to corrector/potentiator treatments. Our results underline the importance to validate treatment efficacy in the context of complex alleles.

The implications of CFTR structural studies for cystic fibrosis drug development.

Development of Cystic Fibrosis Transmembrane conductance Regulator (CFTR) modulators, targeting the root cause of cystic fibrosis (CF), represents a challenge in the era of personalized medicine, as CFTR mutations lead to a variety of phenotypes, which likely require different, specific treatments. CF drug development is also complicated by the need to preserve the right balance between stability and flexibility, required for optimal function of the CFTR protein. In this review, we highlight how structural data can be exploited in this context to understand the molecular mechanisms of disease-associated mutations, to characterize the mechanisms of action of known modulators and to rationalize the search for novel, specific compounds.

Molecular modelling and molecular dynamics of CFTR.

The cystic fibrosis transmembrane conductance regulator (CFTR) protein is a member of the ATP-binding cassette (ABC) transporter superfamily that functions as an ATP-gated channel. Considerable progress has been made over the last years in the understanding of the molecular basis of the CFTR functions, as well as dysfunctions causing the common genetic disease cystic fibrosis (CF). This review provides a global overview of the theoretical studies that have been performed so far, especially molecular modelling and molecular dynamics (MD) simulations. A special emphasis is placed on the CFTR-specific evolution of an ABC transporter framework towards a channel function, as well as on the understanding of the effects of disease-causing mutations and their specific modulation. This in silico work should help structure-based drug discovery and design, with a view to develop CFTR-specific pharmacotherapeutic approaches for the treatment of CF in the context of precision medicine.

New insights into interactions between the nucleotide-binding domain of CFTR and keratin 8.

The intermediate filament protein keratin 8 (K8) interacts with the nucleotide-binding domain 1 (NBD1) of the cystic fibrosis (CF) transmembrane regulator (CFTR) with phenylalanine 508 deletion (ΔF508), and this interaction hampers the biogenesis of functional ΔF508-CFTR and its insertion into the plasma membrane. Interruption of this interaction may constitute a new therapeutic target for CF patients bearing the ΔF508 mutation. Here, we aimed to determine the binding surface between these two proteins, to facilitate the design of the interaction inhibitors. To identify the NBD1 fragments perturbed by the ΔF508 mutation, we used hydrogen-deuterium exchange coupled with mass spectrometry (HDX-MS) on recombinant wild-type (wt) NBD1 and ΔF508-NBD1 of CFTR. We then performed the same analysis in the presence of a peptide from the K8 head domain, and extended this investigation using bioinformatics procedures and surface plasmon resonance, which revealed regions affected by the peptide binding in both wt-NBD1 and ΔF508-NBD1. Finally, we performed HDX-MS analysis of the NBD1 molecules and full-length K8, revealing hydrogen-bonding network changes accompanying complex formation. In conclusion, we have localized a region in the head segment of K8 that participates in its binding to NBD1. Our data also confirm the stronger binding of K8 to ΔF508-NBD1, which is supported by an additional binding site located in the vicinity of the ΔF508 mutation in NBD1.

Functional defect of variants in the adenosine triphosphate-binding sites of ABCB4 and their rescue by the cystic fibrosis transmembrane conductance regulator potentiator, ivacaftor (VX-770).

ABCB4 (MDR3) is an adenosine triphosphate (ATP)-binding cassette (ABC) transporter expressed at the canalicular membrane of hepatocytes, where it mediates phosphatidylcholine (PC) secretion. Variations in the ABCB4 gene are responsible for several biliary diseases, including progressive familial intrahepatic cholestasis type 3 (PFIC3), a rare disease that can be lethal in the absence of liver transplantation. In this study, we investigated the effect and potential rescue of ABCB4 missense variations that reside in the highly conserved motifs of ABC transporters, involved in ATP binding. Five disease-causing variations in these motifs have been identified in ABCB4 (G535D, G536R, S1076C, S1176L, and G1178S), three of which are homologous to the gating mutations of cystic fibrosis transmembrane conductance regulator (CFTR or ABCC7; i.e., G551D, S1251N, and G1349D), that were previously shown to be function defective and corrected by ivacaftor (VX-770; Kalydeco), a clinically approved CFTR potentiator. Three-dimensional structural modeling predicted that all five ABCB4 variants would disrupt critical interactions in the binding of ATP and thereby impair ATP-induced nucleotide-binding domain dimerization and ABCB4 function. This prediction was confirmed by expression in cell models, which showed that the ABCB4 mutants were normally processed and targeted to the plasma membrane, whereas their PC secretion activity was dramatically decreased. As also hypothesized on the basis of molecular modeling, PC secretion activity of the mutants was rescued by the CFTR potentiator, ivacaftor (VX-770). CONCLUSION: Disease-causing variations in the ATP-binding sites of ABCB4 cause defects in PC secretion, which can be rescued by ivacaftor. These results provide the first experimental evidence that ivacaftor is a potential therapy for selected patients who harbor mutations in the ATP-binding sites of ABCB4. (Hepatology 2017;65:560-570).

Pseudomonas aeruginosa Expresses a Functional Human Natriuretic Peptide Receptor Ortholog: Involvement in Biofilm Formation.

UNLABELLED: Considerable evidence exists that bacteria detect eukaryotic communication molecules and modify their virulence accordingly. In previous studies, it has been demonstrated that the increasingly antibiotic-resistant pathogen Pseudomonas aeruginosa can detect the human hormones brain natriuretic peptide (BNP) and C-type natriuretic peptide (CNP) at micromolar concentrations. In response, the bacterium modifies its behavior to adapt to the host physiology, increasing its overall virulence. The possibility of identifying the bacterial sensor for these hormones and interfering with this sensing mechanism offers an exciting opportunity to directly affect the infection process. Here, we show that BNP and CNP strongly decrease P. aeruginosa biofilm formation. Isatin, an antagonist of human natriuretic peptide receptors (NPR), prevents this effect. Furthermore, the human NPR-C receptor agonist cANF(4-23) mimics the effects of natriuretic peptides on P. aeruginosa, while sANP, the NPR-A receptor agonist, appears to be weakly active. We show in silico that NPR-C, a preferential CNP receptor, and the P. aeruginosa protein AmiC have similar three-dimensional (3D) structures and that both CNP and isatin bind to AmiC. We demonstrate that CNP acts as an AmiC agonist, enhancing the expression of the ami operon in P. aeruginosa. Binding of CNP and NPR-C agonists to AmiC was confirmed by microscale thermophoresis. Finally, using an amiC mutant strain, we demonstrated that AmiC is essential for CNP effects on biofilm formation. In conclusion, the AmiC bacterial sensor possesses structural and pharmacological profiles similar to those of the human NPR-C receptor and appears to be a bacterial receptor for human hormones that enables P. aeruginosa to modulate biofilm expression. IMPORTANCE: The bacterium Pseudomonas aeruginosa is a highly dangerous opportunist pathogen for immunocompromised hosts, especially cystic fibrosis patients. The sites of P. aeruginosa infection are varied, with predominance in the human lung, in which bacteria are in contact with host molecular messengers such as hormones. The C-type natriuretic peptide (CNP), a hormone produced by lung cells, has been described as a bacterial virulence enhancer. In this study, we showed that the CNP hormone counteracts P. aeruginosa biofilm formation and we identified the bacterial protein AmiC as the sensor involved in the CNP effects. We showed that AmiC could bind specifically CNP. These results show for the first time that a human hormone could be sensed by bacteria through a specific protein, which is an ortholog of the human receptor NPR-C. The bacterium would be able to modify its lifestyle by favoring virulence factor production while reducing biofilm formation.

Full-open and closed CFTR channels, with lateral tunnels from the cytoplasm and an alternative position of the F508 region, as revealed by molecular dynamics.

In absence of experimental 3D structures, several homology models, based on ABC exporter 3D structures, have provided significant insights into the molecular mechanisms underlying the function of the cystic fibrosis transmembrane conductance regulator (CFTR) protein, a chloride channel whose defects are associated with cystic fibrosis (CF). Until now, these models, however, did not furnished much insights into the continuous way that ions could follow from the cytosol to the extracellular milieu in the open form of the channel. Here, we have built a refined model of CFTR, based on the outward-facing Sav1866 experimental 3D structure and integrating the evolutionary and structural information available today. Molecular dynamics simulations revealed significant conformational changes, resulting in a full-open channel, accessible from the cytosol through lateral tunnels displayed in the long intracellular loops (ICLs). At the same time, the region of nucleotide-binding domain 1 in contact with one of the ICLs and carrying amino acid F508, the deletion of which is the most common CF-causing mutation, was found to adopt an alternative but stable position. Then, in a second step, this first stable full-open conformation evolved toward another stable state, in which only a limited displacement of the upper part of the transmembrane helices leads to a closure of the channel, in a conformation very close to that adopted by the Atm1 ABC exporter, in an inward-facing conformation. These models, supported by experimental data, provide significant new insights into the CFTR structure-function relationships and into the possible impact of CF-causing mutations.

Human genetic polymorphisms in T1R1 and T1R3 taste receptor subunits affect their function.

Umami is the typical taste induced by monosodium glutamate (MSG), which is thought to be detected by the heterodimeric G protein-coupled receptor, T1R1 and T1R3. Previously, we showed that MSG detection thresholds differ substantially between individuals and we further showed that nontaster and hypotaster subjects are associated with nonsynonymous single polymorphisms occurring in the T1R1 and T1R3 genes. Here, we show using functional expression that both amino acid substitutions (A110V and R507Q) in the N-terminal ligand-binding domain of T1R1 and the 2 other ones (F749S and R757C), located in the transmembrane domain of T1R3, severely impair in vitro T1R1/T1R3 response to MSG. A molecular model of the ligand-binding region of T1R1/T1R3 provides a mechanistic explanation supporting functional expression data. The data presented here support causal relations between the genotype and previous in vivo psychophysical studies in human evaluating sensitivity to MSG.

A new protein binding pocket similarity measure based on comparison of clouds of atoms in 3D: application to ligand prediction.

BACKGROUND: Predicting which molecules can bind to a given binding site of a protein with known 3D structure is important to decipher the protein function, and useful in drug design. A classical assumption in structural biology is that proteins with similar 3D structures have related molecular functions, and therefore may bind similar ligands. However, proteins that do not display any overall sequence or structure similarity may also bind similar ligands if they contain similar binding sites. Quantitatively assessing the similarity between binding sites may therefore be useful to propose new ligands for a given pocket, based on those known for similar pockets. RESULTS: We propose a new method to quantify the similarity between binding pockets, and explore its relevance for ligand prediction. We represent each pocket by a cloud of atoms, and assess the similarity between two pockets by aligning their atoms in the 3D space and comparing the resulting configurations with a convolution kernel. Pocket alignment and comparison is possible even when the corresponding proteins share no sequence or overall structure similarities. In order to predict ligands for a given target pocket, we compare it to an ensemble of pockets with known ligands to identify the most similar pockets. We discuss two criteria to evaluate the performance of a binding pocket similarity measure in the context of ligand prediction, namely, area under ROC curve (AUC scores) and classification based scores. We show that the latter is better suited to evaluate the methods with respect to ligand prediction, and demonstrate the relevance of our new binding site similarity compared to existing similarity measures. CONCLUSIONS: This study demonstrates the relevance of the proposed method to identify ligands binding to known binding pockets. We also provide a new benchmark for future work in this field. The new method and the benchmark are available at http://cbio.ensmp.fr/paris/.

Virtual screening of GPCRs: an in silico chemogenomics approach.

BACKGROUND: The G-protein coupled receptor (GPCR) superfamily is currently the largest class of therapeutic targets. In silico prediction of interactions between GPCRs and small molecules in the transmembrane ligand-binding site is therefore a crucial step in the drug discovery process, which remains a daunting task due to the difficulty to characterize the 3D structure of most GPCRs, and to the limited amount of known ligands for some members of the superfamily. Chemogenomics, which attempts to characterize interactions between all members of a target class and all small molecules simultaneously, has recently been proposed as an interesting alternative to traditional docking or ligand-based virtual screening strategies. RESULTS: We show that interaction prediction in the chemogenomics framework outperforms state-of-the-art individual ligand-based methods in accuracy both for receptor with known ligands and without known ligands. This is done with no knowledge of the receptor 3D structure. In particular we are able to predict ligands of orphan GPCRs with an estimated accuracy of 78.1%. CONCLUSION: We propose new methods for in silico chemogenomics and validate them on the virtual screening of GPCRs. The methods represent an extension of a recently proposed machine learning strategy, based on support vector machines (SVM), which provides a flexible framework to incorporate various information sources on the biological space of targets and on the chemical space of small molecules. We investigate the use of 2D and 3D descriptors for small molecules, and test a variety of descriptors for GPCRs. We show that incorporating information about the known hierarchical classification of the target family and about key residues in their inferred binding pockets significantly improves the prediction accuracy of our model.