The de novo drug design based on SMILES format is a typical sequence-processing problem. Previous methods based on recurrent neural network (RNN) exhibit limitation in capturing long-range dependency, resulting in a high invalid percentage in generated molecules. Recent studies have shown the potential of Transformer architecture to increase the capacity of handling sequence data. In this work, the encoder module in the Transformer is used to build a generative model. First, we train a Transformer-encoder-based generative model to learn the grammatical rules of known drug molecules and a predictive model to predict the activity of the molecules. Subsequently, transfer learning and reinforcement learning were used to fine-tune and optimize the generative model, respectively, to design new molecules with desirable activity. Compared with previous RNN-based methods, our method has improved the percentage of generating chemically valid molecules (from 95.6 to 98.2%), the structural diversity of the generated molecules, and the feasibility of molecular synthesis. The pipeline is validated by designing inhibitors against the human BRAF protein. Molecular docking and binding mode analysis showed that our method can generate small molecules with higher activity than those carrying ligands in the crystal structure and have similar interaction sites with these ligands, which can provide new ideas and suggestions for pharmaceutical chemists.

Safe and effective drugs require a favorable combination of pharmacodynamics and pharmacokinetics.

In the development of targeted drugs, anticancer peptides (ACPs) have attracted great attention because of their high selectivity, low toxicity and minimal non-specificity. In this work, we report a framework of ACPs generation, which combines Wasserstein autoencoder (WAE) generative model and Particle Swarm Optimization (PSO) forward search algorithm guided by attribute predictive model to generate ACPs with desired properties. It is well known that generative models based on Variational AutoEncoder (VAE) and Generative Adversarial Networks (GAN) are difficult to be used for de novo design due to the problems of posterior collapse and difficult convergence of training. Our WAE-based generative model trains more successfully (lower perplexity and reconstruction loss) than both VAE and GAN-based generative models, and the semantic connections in the latent space of WAE accelerate the process of forward controlled generation of PSO, while VAE fails to capture this feature. Finally, we validated our pipeline on breast cancer targets (HIF-1) and lung cancer targets (VEGR, ErbB2), respectively. By peptide-protein docking, we found candidate compounds with the same binding sites as the peptides carried in the crystal structure but with higher binding affinity and novel structures, which may be potent antagonists that interfere with these target-mediated signaling.

Metronidazole is a specific drug against trichomonas and anaerobic bacteria, and is widely used in the clinic. However, extensive clinical application is often accompanied by extensive side effects, so it is still of great significance to develop metronidazole derivatives with a new skeleton. Compared with other traditional receptor-based drug design methods, the computational model based on a neural network has higher accuracy and reliability. In this work, a Recurrent Neural Network (RNN) model is applied to the discovery of metronidazole drugs with a new skeleton. Firstly, the generation model based on a Gated Recurrent Unit (GRU) is trained to generate an effective Simplified Molecular-Input Line-Entry System (SMILES) string library with high precision. Then, transfer learning is introduced to fine-tune the GRU model, and many molecules with structures similar to known active drugs are generated. After cluster analysis of the structures of the new compounds, 20 small molecular compounds with metronidazole structures of all different categories were selected, of which 19 may not belong to any published patents or applications. Through prediction and personal experience, the difficulty of synthesizing these 20 new structures was analyzed, and compound 0001 was chosen as our synthetic target, and a series of structures (8a–l) similar to compound 0001 were synthesized. Finally, the inhibitory activities of these compounds against bacteria E. coli, P. aeruginosa, B. subtilis and S. aureus were determined. The results showed that compound 8a–l had obvious inhibitory activity against these four bacteria, which proved the accuracy of our compound generation model.

Extracellular-regulated kinase (ERK2) has been regarded as an essential target for various cancers, especially melanoma. Recently, pyrrolidine piperidine derivatives were reported as Type I^1/2 inhibitors of ERK2, which occupy both the ATP binding pocket and the allosteric pocket. Due to the dynamic behavior of ERK2 upon the binding of Type I^1/2 inhibitors, it is difficult to predict the binding structures and relative binding potencies of these inhibitors with ERK2 accurately. In this work, the binding mechanism of pyrrolidine piperidines was discussed by using different simulation techniques, including molecular docking, ensemble docking based on multiple receptor conformation, molecular dynamics simulations and free energy calculations. Our computational results show that the traditional docking method cannot predict the relative binding ability of the studied inhibitors with high accuracy, but incorporating ERK2 protein flexibility into docking is an effective method to improve the prediction accuracy. It is worth noting that the binding free energies predicted by MM/GBSA or MM/PBSA based on the MD simulations for the docked poses have the highest correlation with the experimental data, which highlights the importance of protein flexibility for accurately predicting the binding ability of Type I^1/2 inhibitors of ERK2. In addition, the comprehensive analysis of several representative inhibitors indicates that hydrogen bonds and hydrophobic interactions are of significance for improving the binding affinities of the inhibitors. We hope this work will provide valuable information for further design of novel and efficient Type I^1/2 ERK2 inhibitors.

NLRP3 (Nucleotide-binding oligomerization domain, LRR and pyrin domain-containing protein 3) is a pivotal regulator of inflammation, with strong implications in gout, neurodegenerative diseases, and various inflammatory conditions. Consequently, the exploration of NLRP3 inhibitors is of great significance for the treatment of diseases. MCC950, NP3-146, compound (3), and YQ128 are four highly bioactive NLRP3 inhibitors that show great potential; however, their mechanism of action is currently limited to targeting the ATP binding region (NACHT site) of the NLRP3 protein. To gain deeper insights into the defining features of NLRP3 inhibitors and to develop more potent inhibitors, it is imperative to elucidate the interaction mechanism between NLRP3 and these inhibitors. In this study, we employ a comprehensive computational approach to investigate the binding mechanism between NLRP3 and representative inhibitors. Utilizing the molecular mechanics/generalized Born surface area (MM/GBSA) method, we calculate the binding free energy and pinpoint the key residues involved in the binding of the four inhibitors to NLRP3. The decomposition of binding free energy by the MM/GBSA method reveals that the residues Val71, Arg195, Ile255, Phe419, Arg422, and Met505, situated around the binding pocket, play a crucial role in conferring the high bioactivity of NLRP3 inhibitors. Furthermore, pharmacophore analysis of the four NLRP3 complexes indicates that the primary interaction between the inhibitors and NLRP3 was mainly hydrophobic interaction. Our study provides a profound understanding of the interaction mechanism between NLRP3 and its inhibitors, identifies the key residues involved, and provides theoretical guidance for the design of more efficient NLRP3 inhibitors.