Journal of Chemical Information and Modeling, Vol. 66, pp. 6131-6140
Abstract: Protein-protein interactions (PPI) play a crucial role in nearly all cellular processes, and their dysregulation often leads to diseases. Stabilizing rather than inhibiting PPIs by small drug-like molecules offers a promising route to modulate PPIs. Here, we present an effective workflow (PPIS-MDPharma) to identify PPI stabilizers solely from molecular dynamic (MD) simulation trajectories of protein-protein (PP) complexes in the absence of a stabilizer and large database pharmacophore screening. Our approach involves extracting pharmacophore features, namely, hydrogen bonding, electrostatic, hydrophobic, and aromatic features from MD simulation by analyzing the interaction of the interface pocket residues with water and ions. The resulting pharmacophore model, along with tens of thousands of derived subsets, is ranked and screened against a local database of 50 million compounds using rapid pharmacophore screening. It yields tens of thousands of stabilizer candidates followed by rescoring using the molecular mechanics generalized Born surface area (MMGBSA) method. For seven PP complexes, the top-ranked ligands exhibited MMGBSA scores similar to experimentally known stabilizers. The approach is computationally more efficient than alternative docking based methods, making it a promising tool for discovering novel PPI stabilizers for various therapeutic applications.
Computational and Structural Biotechnology Journal, Vol. 31, pp. 61-73
Abstract: Cathepsins are papain-like proteolytic enzymes localized in lysosomes and the extracellular matrix, where they participate in diverse physiological and pathological processes. They are synthesized as inactive precursors—procathepsins—containing a propeptide domain that blocks access to the active site. The activity of (pro)cathepsins can be modulated by glycosaminoglycans (GAGs), which are negatively charged, sulfated polysaccharides. This study aimed to develop machine learning (ML) models to predict MM-GBSA binding free energies in (pro)cathepsin–GAG complexes. Molecular dynamics simulations were performed using the ff14SB/GLYCAM06j force field for six (pro)cathepsins and six GAGs, representing four periodic states and six binding poses. Structural and energetic descriptors derived from these simulations were used as input features for eight ML algorithms: ElasticNet, Linear Regression, LinearSVR (with RBFSampler), LightGBM, Histogram Gradient Boosting, Fully Connected Neural Network (FCNN), and Random Forest. The FCNN yielded the most accurate predictions (R2 = 0.7124 ± 0.0089; MAE = 5.2033 ± 0.0876 kcal/mol), with GradientBoost-based models performing comparably. Optimal FCNN performance was achieved with a minimal architecture (no hidden layers, dropout rate 0.01, ReLU activation). Incorporating Linear Interaction Energy (LIE) components significantly improved prediction accuracy, and approximately 17,000 data points were sufficient for stable model performance. Overall, this study provides a proof of concept for using ML to estimate binding free energies in protein–GAG systems and establishes a foundation for generalizable, structure-based predictors applicable to a broad range of biomolecular complexes. Beyond predictive accuracy, this approach enables rapid screening of MMGBSA interactions, facilitating the identification of favorable binding regions and accelerating structure-guided design efforts.
Journal of chemical information and modeling, Vol. 66, pp. 6159-6180
Abstract: Glycosaminoglycans (GAGs) are long, anionic polysaccharides abundant in the extracellular matrix and lysosomes, where their electrostatic interactions with proteins are essential for biological function. Computational studies of GAG-containing systems remain challenging due to their significant charge density and conformational flexibility. Here we benchmark two widely used force-fields, ff14SB/GLYCAM06j-1 and CHARMM36m, for three experimentally characterized protein-GAG complexes. Both force fields reproduce the key structural features of protein-GAG interactions, while GAG dynamics depend on protein charge, with CHARMM36m favoring broader surface exploration for highly positively charged proteins and AMBER enhancing mobility for less charged systems. Although protein flexibility is similarly described, ff14SB/GLYCAM06j-1 samples a broader GAG conformational space, and dissociation free energy profiles diverge for highly anionic GAGs, but remain comparable for moderately sulfated systems. In addition, we performed molecular dynamics simulations for all systems using the ff14SB/GLYCAM06j-1, CHARMM36m, and ff19SB/GLYCAM06j-1 force fields in a 15 Å solvent box. Structural and energetic analyses revealed no significant impact of the solvent box size on the examined descriptors. These results establish practical benchmarks for accurate atomistic simulations of GAG-protein assemblies and will inform future developments in biomolecular force fields.
Nature Communications , Vol. 17
Abstract: During eukaryotic DNA replication initiation, inactive MCM2–7 double-hexamers assembled at replication origins must be converted into two active CMG helicases, yet how this transition is coupled to origin DNA unwinding in vivo remains unclear. Here, we identify a DNA-bound intermediate with an extended genomic footprint that forms during helicase activation. Genome-wide mapping of initial strand separation reveals that DNA unwinding initiates near the N-terminal interface of opposing MCM2–7 hexamers. At these sites, the origin DNA exhibits a conserved AT-rich/GC-rich/AT-rich sequence architecture centred under the helicase complex, which is associated with an elevated DNA melting probability. We further show that restricting hexamer splitting delays release of the Cdc45-loading factor Sld3, demonstrating that mechanical transitions during helicase activation are tightly coupled to complex disassembly. Finally, we provide in vivo evidence that single-stranded DNA is ejected through a specialised DNA exit gate at the Mcm2/5 interface during helicase activation, which is dispensable for ongoing DNA synthesis. Together, these findings establish a mechanistic framework for how replication origins are remodelled to initiate DNA replication and reveal key intermediates and DNA transactions during helicase activation.
Virchows Archiv, Vol. 489, pp. 219-224
Abstract: Small cell lung carcinoma (SCLC) is classically defined by biallelic inactivation of RB1 and TP53. However, a small subset of tumors retains Rb expression and exhibits distinct molecular features. Here, we report two Rb-retained SCLC cases that expand the biological and therapeutic spectrum of this subgroup. Both tumors occurred in middle-aged women, showed small cell morphology with some variant features, and displayed complex copy number alterations. Case 1 harbored a truncal KRAS p.G12C mutation with high-level amplification of chromosome 11q13-q14, including CCND1, and demonstrated a clinical response to sotorasib. Case 2 harbored a TP53 mutation, CDKN2A loss, STK11 inactivation, and a novel IKZF2::ERBB4 fusion. These findings highlight the molecular heterogeneity of Rb-retained SCLC and demonstrate that this subgroup can harbor clinically actionable oncogenic drivers. Accordingly, routine assessment of Rb expression in SCLC, followed by comprehensive molecular profiling of Rb-retained tumors, is warranted to uncover therapeutically relevant targets.
Proteins: Structure, Function and Bioinformatics, Vol. 94, pp. 1356-1370
Abstract: The control and modulation of protein–protein interactions (PPIs) is of central importance for the majority of biological processes and most biomedical applications. Stabilization of PPIs, besides inhibition, is of growing pharmaceutical interest. Due to their small size, drug-like organic molecules may not provide sufficient interaction surfaces to allow for high-affinity dual binding to both partners of a protein–protein complex. Cyclic peptides offer larger interaction surfaces, making them a promising class of stabilizers of PPIs. We have developed a computational protocol to rapidly and systematically design cyclic peptides that optimize not only the interaction with one target protein but simultaneously optimize the dual binding to two protein partners. The cyclic peptide generation is based on a modified AlphaFold2-based peptide design approach and combines confidence scoring with force field-based scoring using Molecular Dynamics simulations. The performance of the method is tested on protein–protein complexes with known cyclic peptide binders and stabilizers. In addition, the approach is used to design cyclic peptides that can act as bifunctional molecules, recruiting the cellular protein degradation system to a target protein. The designed cyclic peptides achieve similar or better calculated interaction scores than known binders and exhibit well-balanced interactions with both protein partners. The design protocol is generally applicable to cyclic peptide design for modulating or inducing protein–protein association and could be useful for many biomedical design efforts.
Journal of Molecular Biology, Vol. 438
Abstract: Cellular metabolic systems but also the extracellular environment can generate reactive oxygen species that lead to oxidation of methionine (MET) and interfere with protein folding and protein–protein association. The molecular mechanism of how MET oxidation (MEO) influences conformational stability and binding is not well understood. We employ alchemical free energy simulations to systematically study the influence of MET oxidation on protein–protein binding using the tetramerization domain of the tumor suppression protein p53 as a model system. A single MEO in one tetramerisation domain destabilizes the tetramer by ≈1.1–1.8 kcal/mol depending slightly on the MEO diastereomer. The simulations on double and triple oxidations reveal increased destabilization (≈3–7 kcal/mol) and significant cooperative effects depending on the relative position of the oxidized residues. The MET oxidation effects are of similar magnitude for the change in stability of the human prion protein (HPP) that served as a second model system and also agreed with available experimental data. The calculations predict a significant dependence of stability changes on the position of the MEO and also indicate non-additive effects of multiple oxidations which may play a role to protect proteins from oxidative damage and stress. Analysis of the Molecular Dynamics trajectories allowed us to interpret the oxidation effects in molecular detail. The simulation methodology could also serve as a general protocol to analyze single and multiple MET oxidations in other systems and its influence on protein binding and stability.
Bioinformatics, Vol. 42
Abstract: Motivation The rational design of chemical compounds that bind to a desired protein target molecule is a major goal of drug discovery. Most current molecular docking but also fragment-based buildup or machine learning-based generative drug design approaches employ a rigid protein target structure. Results Based on recent progress in predicting protein structures and complexes with chemical compounds, we have designed an approach, AI-MCLig, to optimize a chemical compound bound to a fully flexible and conformationally adaptable protein binding region. During a Monte Carlo (MC)-type simulation to randomly change a chemical compound, the target protein–compound complex is completely rebuilt at every MC step using the Chai-1 protein structure prediction program. Besides compound flexibility it allows the protein to adapt to the chemically changing compound. MC protocols based on atom-/bond-type changes or based on combining larger chemical fragments have been tested. Simulations on four test targets resulted in potential ligands that show very good binding scores comparable to experimentally known binders using several different scoring schemes. The MC-based compound design approach is complementary to existing approaches and could help for the rapid design of putative binders including induced fit of the protein target. Availability and implementation Datasets, examples, and source code are available on our public GitHub repository https://github.com/JakobAgamia/AI-MCLig and on Zenodo at https://doi.org/10.5281/zenodo.17800140.
Chemical Science, Vol. 17, pp. 3198-3211
Abstract: DNA mimic foldamers are helically folded aromatic oligoamides bearing negatively charged side chains that mimic the shape and charge distribution of double-stranded B-DNA. They have been shown to bind to some DNA-binding proteins better than DNA itself and thus have potential to interfere with DNA-protein interactions. Their structure has been previously characterized in detail by X-ray crystallography. We have now investigated their structural dynamics both computationally and experimentally. The force field parameters of the building blocks required for DNA mimicry were optimized and implemented in AMBER to perform molecular dynamics simulations of the foldamer helices. The position of the negatively charged side chains on the helix, the charge state of the side chains, and the presence of salt were systematically varied. The simulations revealed that the global flexibility parameters for twisting and bending of the foldamer helices are of similar magnitude to those of B-DNA, though distinct kinking events and motions are involved. A range of sequences were then prepared for experimental investigations using 1H NMR, UV-vis absorption and circular dichroism spectroscopies. Measurements revealed that the foldamer helices are stable over a broad range of temperature, pH and salt conditions in aqueous solutions, but that they nevertheless undergo structural changes when conditions are modified. An assay was developed to quantitatively assess foldamer helix stability through the measurement of the rate of interconversion between right-handed and left-handed diastereomeric conformers. Unexpectedly, suppressing some negatively charged side chains had a destabilizing effect on the helix, suggesting a more complex role of the side chains than electrostatic repulsions.
Computational Modeling of Biomolecular Interactions, pp. 213-242
Abstract: Great progress has been achieved in recent years to accurately model the structure of protein-protein, protein-peptide, and also protein-RNA complexes. Without doubt, recent artificial intelligence-based deep learning prediction approaches have achieved breakthroughs in modeling such complexes often with unprecedented accuracy. However, traditional docking methods and molecular dynamics simulation approaches provide useful extension for inclusion of experimental data and for obtaining thermodynamic and kinetic insights into biomolecular association processes. An overview will be given on traditional molecular docking methods, molecular dynamics simulation, and advanced sampling methods. Recent AI-based prediction methods and discussions into how such approaches can be combined with traditional docking and simulation methods to obtain new insights into the molecular mechanisms of protein binding to partner proteins and other types of biomolecules are also provided.
