FFLOM: A Flow-Based Autoregressive Model for Fragment-to-Lead Optimization.

Recently, deep generative models have been regarded as promising tools in fragment-based drug design (FBDD). Despite the growing interest in these models, they still face challenges in generating molecules with desired properties in low data regimes. In this study, we propose a novel flow-based autoregressive model named FFLOM for linker and R-group design. In a large-scale benchmark evaluation on ZINC, CASF, and PDBbind test sets, FFLOM achieves state-of-the-art performance in terms of validity, uniqueness, novelty, and recovery of the generated molecules and can recover over 92% of the original molecules in the PDBbind test set (with at least five atoms). FFLOM also exhibits excellent potential applicability in several practical scenarios encompassing fragment linking, PROTAC design, R-group growing, and R-group optimization. In all four cases, FFLOM can perfectly reconstruct the ground-truth compounds and generate over 74% of molecules with novel fragments, some of which have higher binding affinity than the ground truth.

Discovery of novel inhibitors of CDK2 using docking and physics-based binding free energy calculation.

Cyclin-dependent kinase (CDK) is a serine/threonine protein kinase family that cooperates with cyclin and plays an important role in the regulation of cell cycle. Cyclin-dependent kinase 2 is an important member of the CDK family and holds great promise as an anti-cancer drug target. In this study, we used molecular docking and physics-based binding free energy calculation method AS-IE that explicitly calculated protein-ligand binding entropy to discover novel inhibitors of CDK2. A total of 17 inhibitors were discovered with the best IC50 reaching ~2 µM. Decomposition of the binding free energy using AS-IE reveals key protein-ligand interactions that determines the activity. These results provided a good example of drug design using physics-based free energy calculation method such as AS-IE and the novel compounds offered a good start point for further development of CDK2 inhibitors.

Insights into small molecule inhibitor bindings to PD-L1 with residue-specific binding free energy calculation.

Targeting the immunological checkpoint PD-1/PD-L1 with antibodies has shown opportunities to improve cancer treatment in recent years. However, antibody therapy is a double-edged sword with high cost, low patient tolerance, lack of oral bioavailability, and a reaction to most solid tumors that prevents the adoption of antibodies. Advancement of small-molecule PD-1/PD-L1 inhibitors that could overwhelm these drawbacks is sluggish because of the poor pharmacodynamic properties and shallow pocket of the PD-1/PD-L1 binding interface. Recently, a number of compounds have been discovered to bind the PD-L1/PD-L1 dimer interface, providing an excellent alternative to inhibit the interaction between PD-1/PD-L1 and small molecules. Quantitative characterization of PD-L1 interactions with these inhibitors will advance the design of novel and efficient inhibitors in the future. Here, the binding free energies of 35 PD-L1 dimer inhibitors have been calculated using the alanine-scanning-interaction-entropy (AS-IE) method. Hotspot residues on PD-L1 and potential modification groups on the inhibitors were identified. The experimental results for the AS-IE method were better correlated than the classical MM/GBSA method. These results may set the stage for the design the more powerful PD-L1 inhibitors.Communicated by Ramaswamy H. Sarma.

Discovery of novel inhibitors of SARS-CoV-2 main protease.

Corona Virus Disease 2019 (COVID-19), referred to as 'New Coronary Pneumonia', is a type of acute infectious disease caused by the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection. Mpro is one of the main targets for treating COVID-19. The current research on Mpro mainly focuses on the repurposing of old drugs, and there are only a few novel ligands that inhibit Mpro. In this research, we used computational free energy calculation to screen a compound library against Mpro, and discovered four novel compounds with the two best compounds (AG-690/13507628 and AG-690/13507724) having experimental measured IC50 of just under 3 µM and low cell toxicity. Detailed decomposition of the interactions between the inhibitors and Mpro reveals key interacting residues and interactions that determine the activity. The results from this study should provide a basis for further development of anti-SARS-CoV-2 drugs.Communicated by Ramaswamy H. Sarma.

Residue-specific binding mechanisms of PD-L1 to its monoclonal antibodies by computational alanine scanning.

Programmed cell death 1 receptor (PD-1) on the surface of T cells and its ligand 1 (PD-L1) are immune checkpoint proteins. Treating cancer patients with inhibitors blocking this checkpoint has significantly prolonged the survival rate of patients. In this study, we examined several monoclonal antibodies (mAbs) of PD-L1 and studied their detailed binding mechanism to PD-L1. An efficient computational alanine scanning method was used to perform quantitative analysis of hotspot residues that are important for PD-1/PD-L1 binding. A total of five PD-L1/mAb complexes were investigated and hotspots on both PD-L1 and mAbs were predicted. Our result shows that PD-L1M115 and PD-L1Y123 are two relatively important hotspots in all the five PD-L1/mAb binding complexes. It is also found that the important residues of mAbs binding to PD-L1M115 and PD-L1Y123 are similar to each other. The computational alanine scanning result is compared to the experimental measurements that are available for two of the mAbs (KN035 and atezolizumab). The calculated alanine scanning result is in good agreement with the experimental data with a correlation coefficient of 0.87 for PD-L1/KN035 and 0.6 for PD-L1/atezolizumab. Our computation found more hotspots on PD-L1 in the PD-L1/KN035 complex than those in the PD-L1/atezolizumab system, indicating stronger binding affinity in the former than the latter, which is in good agreement with the experimental finding. The present work provides important insights for the design of new mAbs targeting PD-L1.

DeepBSP-a Machine Learning Method for Accurate Prediction of Protein-Ligand Docking Structures.

In recent years, machine-learning-based scoring functions have significantly improved the scoring power. However, many of these methods do not perform well in distinguishing the native structure from docked decoy poses due to the lack of decoy structural information in their training data. Here, we developed a machine-learning model, named DeepBSP, that can directly predict the root mean square deviation (rmsd) of a ligand docking pose with reference to its native binding pose. Unlike the binding affinity, the rmsd between the docking poses with reference to their native structures can be straightforwardly determined. By training on a generated data set with 11,925 native complexes and more than 165,000 docked poses, our model shows excellent docking power on our test set and also on the CASF-2016 docking decoy set compared to other major scoring functions. Thus, by combining molecular dockings that generate many poses with the application of DeepBSP, one can more accurately predict the best binding pose that is closest to the native complex structure. This DeepBSP model shall be very useful in picking out poses close to their natives from many poses generated from a dock application.

Analysis of the binding modes of the first- and second-generation antiandrogens with respect to F876L mutation.

Androgen receptor (AR) is an important target for the treatment of prostate cancer, and mutations in the AR have an important impact on the resistance of existing drugs. In this work, we performed molecular dynamics simulations of the existing marketed antiandrogens flutamide, nilutamide, bicalutamide, enzalutamide, apalutamide, darolutamide, and its main metabolite ORM15341 in complex with the wild-type and F876L mutant AR. We calculated the residue-specific binding free energy contribution of the wild-type and mutant ARs with the AS-IE method and analyzed the hotspot residues and the binding free energy contributions of specific residues before and after the mutation. In addition, we analyzed the total binding obtained by adding residue binding energy contributions and compared the results with experimental values. The obtained residue-specific binding information should be very helpful in understanding the mechanism of drug resistance with respect to specific mutations and in the design of new generation drugs against possible new mutations.

Computational Analysis of Residue-Specific Binding Free Energies of Androgen Receptor to Ligands.

Androgen receptor (AR) is an important therapeutic target for the treatment of diseases such as prostate cancer, hypogonadism, muscle wasting, etc. In this study, the complex structures of the AR ligand-binding domain (LBD) with fifteen ligands were analyzed by molecular dynamics simulations combined with the alanine-scanning-interaction-entropy method (ASIE). The quantitative free energy contributions of the pocket residues were obtained and hotspot residues are quantitatively identified. Our calculation shows that that these hotspot residues are predominantly hydrophobic and their interactions with binding ligands are mainly van der Waals interactions. The total binding free energies obtained by summing over binding contributions by individual residues are in good correlation with the experimental binding data. The current quantitative analysis of binding mechanism of AR to ligands provides important insight on the design of future inhibitors.

Development of a New Scoring Function for Virtual Screening: APBScore.

In this study, we developed a new physical-based scoring function, Atom Pair-Based Scoring function (APBScore), which includes pairwise van der Waals (VDW), electrostatic interaction, and hydrogen bond energies between the receptor and ligand. Despite the simple form of this scoring function, the tests of APBScore on several benchmark datasets show its excellent performance in scoring as compared to other widely used traditional scoring functions. Particularly, the scoring performance of APBScore is among the top-ranking scoring functions for complexes with zinc/ligand interactions. In addition to the scoring power, APBScore also shows good performance in ranking and docking as compared to some traditional scoring functions. In addition, the APBScore is sensitive to receptor/ligand atomic collisions and therefore can correctly identify decoy complex structures with atomic collisions. These features are the result of optimizing atom-pair VDW interactions, performing structural minimization of the initial structures, and treating zinc/ligand interactions more accurately. The source code of APBScore is available at https://github.com/BaoJingxiao/APBScore.

Computational approaches to studying methylated H4K20 recognition by DNA repair factor 53BP1.

Histone lysine methylation regulates the recruitment of mammalian DNA repair factor 53BP1 to the histone H4 lysine 20 (H4K20), through specific recognition of the tandem Tudor domain of 53BP1. The di- and mono-methylated H4K20 bind to 53BP1 with high affinity, but the non- and tri-methylated H4K20 do not. Here, we develop a new approach to carry out computational study to unravel the binding mechanism of methylated H4K20 by 53BP1 and to compute relative binding affinities of different methylations of H4K20 by 53BP1. First, hot spots in 53BP1 were predicted by computational alanine scanning and aromatic cages formed by W1495, Y1500, Y1502, and Y1523 are found to provide the dominant binding to di- and mono-methylated H4K20 in addition to D1521. Secondly, a de-methylation method is proposed to predict relative binding free energies between 53BP1 and different methylated states of H4K20. Finally, the tri-methylated and non-methylated H4K20/53BP1 complexes are found to be dynamically unstable, explaining the experimental finding that neither can bind to 53BP1. The present work provides an important theoretical basis for our understanding of histone methylations of H4K20 and their recognition mechanism by DNA repair factor 53BP1.

Study of SHMT2 Inhibitors and Their Binding Mechanism by Computational Alanine Scanning.

Mitochondrial serine hydroxymethyl transferase isoform 2 (SHMT2) has attracted increasing attention as a pivotal catalyzing regulator of the serine/glycine pathway in the one-carbon metabolism of cancer cells. However, few inhibitors that target this potential anticancer target have been discovered. Quantitative characterization of the interactions between SHMT2 and its known inhibitors should benefit future discovery of novel inhibitors. In this study, we employed a recently developed alanine-scanning-interaction-entropy method to quantitatively calculate the residue-specific binding free energy of 28 different SHMT2 inhibitors that originate from the same skeleton. Major contributing residues from SHMT2 and chemical groups from the inhibitors were identified, and the binding energy of each residue was quantitatively determined, revealing essential features of the protein-inhibitor interaction. The most important contributing residue is Y105 of the B chain followed by L166 of the A chain. The calculated protein-ligand binding free energies are in good agreement with the experimental results and showed better correlation and smaller errors compared with those obtained using the conventional MM/GBSA with the normal mode method. These results may aid the rational design of more effective SHMT2 inhibitors.

Discovery of Baicalin as NDM-1 inhibitor: Virtual screening, biological evaluation and molecular simulation.

The emergence and worldwide spreads of carbapenemase producing bacteria, especially New Delhi metallo-ß-lactamase (NDM-1), has made a great challenge to treat antibiotics-resistant bacterial infections. It can hydrolyse almost all ß-lactam antibacterials. Unfortunately, there are no clinically useful inhibitors of NDM-1. In this study, structure-based virtual screening method led to the identification of Baicalin as a novel NDM-1 inhibitor. Inhibitory assays showed that Baicalin possessed a good inhibition of NDM-1 with IC50 values of 3.89 ±â€¯1.1 µM and restored the susceptibility of E.coli BL21(DE3)/pET28a-NDM-1 to clinically used ß-lactam antibiotics. Molecular docking and molecular dynamics simulations obtained a complex structure between the relatively stable inhibitor molecule Baicalin and NDM-1 enzyme. The results showed that the carboxyl group in Baicalin directly interacted with the Zn2+ in the active center of the enzyme, and the residues such as Glu152, Gln123, Met67, Trp93 and Phe70 in the enzyme formed hydrogen bonds with Baicalin to further stabilize the complex structure.

Generation of a novel long-acting thymosin alpha1-Fc fusion protein and its efficacy for the inhibition of breast cancer in vivo.

Thymosin alpha1 (Tα1) is a multifunctional polypeptide involved in immunoregulation, which universally exists in various organs. The clinical application, however, is limited because of its short half-life. The Fc domain of human IgG has functional properties, improving the affinity and serum half-life. In this study, we fused Tα1 to Fc domain of human IgG1 for generation of a novel long-acting fusion protein, termed Tα1-Fc. Tα1-Fc was expressed, purified, and identified. The results showed that Tα1-Fc was more potent than Tα1 in inhibiting the growth of 4T1 and MCF-7 breast cancer cells in vivo. Furthermore, in the 4T1 tumor model the mice treated with Tα1-Fc exhibited higher level of CD4 and CD8 cells compared with that of the mice Tα1 treated. The interferon-γ and interleukin-2 level in the Tα1-Fc-treated mice was higher than that in the Tα1-treated mice. Tα1-Fc could also alleviate immunosuppression induced by hydrocortisone. In summary, Tα1-Fc provides a novel potent strategy to recruit immune cells against tumors and enhance the antitumor activity of Tα1.

Novel 3D Structure Based Model for Activity Prediction and Design of Antimicrobial Peptides.

The emergence and worldwide spread of multi-drug resistant bacteria makes an urgent challenge for the development of novel antibacterial agents. A perspective weapon to fight against severe infections caused by drug-resistant microorganisms is antimicrobial peptides (AMPs). AMPs are a diverse class of naturally occurring molecules that are produced as a first line of defense by all multi-cellular organisms. Limited by the number of experimental determinate 3D structure, most of the prediction or classification methods of AMPs were based on 2D descriptors, including sequence, amino acid composition, peptide net charge, hydrophobicity, amphiphilic, etc. Due to the rapid development of structural simulation methods, predicted models of proteins (or peptides) have been successfully applied in structure based drug design, for example as targets of virtual ligand screening. Here, we establish the activity prediction model based on the predicted 3D structure of AMPs molecule. To our knowledge, it is the first report of prediction method based on 3D descriptors of AMPs. Novel AMPs were designed by using the model, and their antibacterial effect was measured by in vitro experiments.

Structural and functional aspects of decorsin and its analog as recognized by integrin αIIbß3.

Decorsin is an antagonist of platelet glycoprotein integrin αIIbß3 on platelets; the protein is 39 amino acids long with three disulfide bridges in its tertiary structure. To demonstrate decorsin's mechanism of action, we applied the computational virtual technique and platelet aggregation inhibition assay, which showed that the flanking amino-acid residues of the Arg-Gly-Asp (RGD) motif play an important role in platelet aggregation. The computational simulations revealed that the RGD motif mainly contributes to the stability of the complex when decorsion interacts with integrin αIIbß3. However, the C-terminal residues, such as 34A→W and 35D→R, was also found to possibly play a key role in their binding structures. Moreover, we produced a decorsin analog (A34W plus D35R decorsin), in which the 34A (alanine) and 35D (aspartic acid) residues were respectively substituted by W (tryptophan) and R (arginine). This isoform was then recombinantly expressed in Escherichia coli. Intriguingly, this mutant type showed higher anti-platelet aggregation activity than the wildtype. Our study may further contribute to finding decorsin mutants with higher anti-platelet aggregation activity.

Computational design, functional analysis and antigenic epitope estimation of a novel hybrid of 12 peptides of hirudin and reteplase.

Cardiovascular and cerebrovascular diseases are leading causes of morbidity and mortality for human beings, and thrombosis is the major risk factor. Thrombolytic therapy has been testified to be the most effective approach to cure thrombosis-related diseases. In clinical treatment, we often adopt a combination therapeutic regimen of both thrombolytic and anticoagulant agents to prevent the recurrence of thrombosis. Thus, a novel hybrid (HV12p-rPA) comprised of the C-terminal 12 residues of hirudin-PA (HV12p) and reteplase (rPA) was designed. The three-dimensional structure of this hybrid was mimicked based on homology modeling and refined with dynamics simulation by utilizing Amber12.0 software. The function of the hybrid was analyzed by structure comparison and the root mean square deviation (RMSD) of Cα atoms between the hybrid and native rPA was calculated. The results showed that HV12p, which was located in the N-terminus of the hybrid, was far from the rPA segment of the hybrid and had no influence on the conformational stability of the rPA domain. The RMSD of Cα atoms of these superimposed proteins was about 40Å, implying that the hybrid had a similar spatial conformation to that of native rPA. Additionally, the antigenic epitopes of the hybrid were predicted by estimations of Hopp-Wood hydrophilicity, Janin accessibility, Zimmermane-Simha polarity, Bhaskaran-Ponnuswamy flexibility, as well as secondary structure analysis and Kolaskar-Tongaonkar antigenicity prediction. The results showed that the most likely antigenic determinants were located at or near regions 148-152, 257-262 and 321-330.

Probing the Effect of Two Heterozygous Mutations in Codon 723 of SLC26A4 on Deafness Phenotype Based on Molecular Dynamics Simulations.

A Chinese family was identified with clinical features of enlarged vestibular aqueduct syndrome (EVAS). The mutational analysis showed that the proband (III-2) had EVAS with bilateral sensorineural hearing loss and carried a rare compound heterozygous mutation of SLC26A4 (IVS7-2A>G, c.2167C>G), which was inherited from the same mutant alleles of IVS7-2A>G heterozygous father and c.2167C>G heterozygous mother. Compared with another confirmed pathogenic biallelic mutation in SLC26A4 (IVS7-2A>G, c.2168A>G), these two biallelic mutations shared one common mutant allele and the same codon of the other mutant allele, but led to different changes of amino acid (p.H723D, p.H723R) and both resulted in the deafness phenotype. Structure-modeling indicated that these two mutant alleles changed the shape of pendrin protein encoded by SLC26A4 with increasing randomness in conformation, and might impair pendrin's ability as an anion transporter. The molecular dynamics simulations also revealed that the stability of mutant pendrins was reduced with increased flexibility of backbone atoms, which was consistent with the structure-modeling results. These evidences indicated that codon 723 was a hot-spot region in SLC26A4 with a significant impact on the structure and function of pendrin, and acted as one of the genetic factors responsible for the development of hearing loss.