Deep interactome learning for de novo drug design

Published: Sept. 19, 2023

De novo drug design aims to generate molecules from scratch that possess specific chemical and pharmacological properties. We present a computational approach utilizing interactome-based deep learning for ligand- structure-based generation of drug-like molecules. This method capitalizes on the unique strengths both graph neural networks language models, offering an alternative need application-specific reinforcement, transfer, or few-shot learning. It allows construction compound libraries tailored bioactivity, synthesizability, structural novelty. In order proactively evaluate interactome framework design, potential new ligands targeting binding site human peroxisome proliferator-activated receptor (PPAR) subtype gamma were generated. The top-ranking designs chemically synthesized biophysically biochemically characterized. Potent PPAR partial agonists identified, demonstrating favorable activity desired selectivity profiles nuclear receptors off-target interactions. Crystal structure determination ligand-receptor complex confirmed anticipated mode. successful outcome positively advocates de application in bioorganic medicinal chemistry, enabling creation innovative bioactive

Language: Английский

---

Graph transformer neural network for chemical reactivity prediction

Published: May 16, 2023

Optimizing the properties of advanced drug candidates can be facilitated by directly introducing certain chemical groups without having to synthesize molecules from scratch. However, their complexity often renders reactivity predictions and synthesis planning challenging. Herein, we introduce a graph transformer neural network (GTNN) approach for computational reaction screening identification substrates suitable late-stage functionalization, taking compound alkylation via Minisci-type chemistry as an example. GTNNs were trained on experimentally generated reactions obtained miniaturized high-throughput experimentation literature data. Trained models prospectively applied predicting 3180 heterocyclic molecules, identifying potential alkylation. All predicted confirmed. Multiple transformations identified each these compounds. Selected hits scaled up, isolated, characterized, delivering 30 novel, suitably functionalized medicinal chemistry. These results positively advocate GTNN prediction in discovery.

Language: Английский

---

G-PLIP: Knowledge graph neural network for structure-free protein-ligand bioactivity prediction

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 5, 2023

Abstract Protein-ligand interaction (PLI) shapes efficacy and safety profiles of small molecule drugs. Existing methods rely on either structural information or resource-intensive computation to predict PLI, making us wonder whether it is possible perform structure-free PLI prediction with low computational cost. Here we show that a light-weight graph neural network (GNN), trained quantitative PLIs number proteins ligands, able the strength unseen PLIs. The model has no direct access protein-ligand complexes. Instead, predictive power provided by encoding entire chemical proteomic space in single heterogeneous graph, encapsulating primary protein sequence, gene expression, protein-protein network, similarities between ligands. novel performs competitively better than structure-aware models. Our observations suggest existing PLI-prediction may be further improved using representation learning techniques embed biological knowledge.

Language: Английский

---

Development of a novelin silicoclassification model to assess reactive metabolite formation in the cysteine trapping assay and investigation of important substructures

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 14, 2024

Abstract Predicting whether a compound can cause drug-induced liver injury (DILI) is difficult due to the complexity of its mechanism. The production reactive metabolites one major causes DILI, particularly idiosyncratic DILI. cysteine trapping assay methods detect which bind microsomes covalently. However, it cumbersome use 35S isotope-labeled for this assay. Therefore, we constructed an in silico classification model predicting positive/negative outcome accelerate drug discovery process. In study, collected 475 compounds (436 in-house and 39 publicly available drugs). Using Message Passing Neural Network (MPNN) Random Forest (RF) with extended connectivity fingerprint (ECFP) 4, built machine learning models predict covalent binding risk compounds. 5-fold cross-validation (CV) hold-out test were evaluated random- time-split trials. Additionally, investigated substructures that contributed positive results through framework MPNN model. random-split dataset, AUC-ROC RF 0.698 0.811 CV, 0.742 0.819 test, respectively. 0.729 0.617 0.625 0.559 restrictively. This result suggests has higher predictivity than dataset. Hence, conclude have better predictive power. Furthermore, most positively consistent previous reports such as propranolol, verapamil, imipramine. new determine assay, namely accurately factors We believe contribute mitigating DILI at early stages discovery.

Language: Английский

---

Development of a Novel In Silico Classification Model to Assess Reactive Metabolite Formation in the Cysteine Trapping Assay and Investigation of Important Substructures

Biomolecules, Journal Year: 2024, Volume and Issue: 14(5), P. 535 - 535

Published: April 30, 2024

Predicting whether a compound can cause drug-induced liver injury (DILI) is difficult due to the complexity of drug mechanism. The cysteine trapping assay method for detecting reactive metabolites that bind microsomes covalently. However, it cumbersome use 35S isotope-labeled this assay. Therefore, we constructed an in silico classification model predicting positive/negative outcome We collected 475 compounds (436 in-house and 39 publicly available drugs) based on experimental data performed study, composition results showed 248 positives 227 negatives. Using Message Passing Neural Network (MPNN) Random Forest (RF) with extended connectivity fingerprint (ECFP) 4, built machine learning models predict covalent binding risk compounds. In time-split dataset, AUC-ROC MPNN RF were 0.625 0.559 hold-out test, restrictively. This result suggests has higher predictivity than dataset. Hence, conclude better predictive power. Furthermore, most substructures contributed positively consistent previous results.

Language: Английский

---