A Novel Scalarized Scaffold Hopping Algorithm with Graph-Based Variational Autoencoder for Discovery of JAK1 Inhibitors

Abstract

We have developed a graph-based Variational Autoencoder with Gaussian Mixture hidden space (GraphGMVAE), a deep learning approach for controllable magnitude of scaffold hopping in generative chemistry. It can effectively and accurately generate molecules from a given reference compound, with excellent scaffold novelty against known molecules in the literature or patents (97.9% are novel scaffolds). Moreover, a pipeline for prioritizing the generated compounds was also proposed to narrow down our validation focus. In this work, GraphGMVAE was validated by rapidly hopping the scaffold from FDA-approved upadacitinib, which is an inhibitor of human Janus kinase 1 (JAK1), to generate more potent molecules with novel chemical scaffolds. Seven compounds were synthesized and tested to be active in biochemical assays. The most potent molecule has 5.0 nM activity against JAK1 kinase, which shows that the GraphGMVAE model can design molecules like how a human expert does but with high efficiency and accuracy.

Molecular Modeling Insights into Upadacitinib Selectivity upon Binding to JAK Protein Family

“…In contrast, JAK2 and JAK3 showed the possibility of the 2,2,2-trifluoroethyl group being located under the glycine loop. This behavior was similar to that of the analog compounds used in this study [12]. There were obstructions created by the imidazole cycle of histidine in the glycine loop during visual inspection of the MD trajectories.…”

Multimodal generative neural networks and molecular dynamics based identification of PDK1 PIF-pocket modulators

“…Recently, deep generative models are successfully utilized in the chemical structure generation of binders for different targets such as discoidin domain receptor 1 (DDR1), 13 Bruton's tyrosine kinase (BTK), 14 D4 dopamine receptor (DRD4), 12 BRAF protein, 15 protease of SARS-CoV-2, 16 human Janus kinase 1 (JAK1), 17 and the receptor for advanced glycation end products (RAGE). 18 The identification of molecules in a deep generative framework can be made by directly generating 1D SMILES, 2D graphs, or 3D ligands.…”

Application advances of deep learning methods for de novo drug design and molecular dynamics simulation

“…[7][8][9] Since deep learning methods can leverage multiple processing layers to automatically extract task-related features, they have made rapid progress in the fields of image classification, 10 speech recognition, 11 natural language processing, drug discovery, and MD simulations, among others. [12][13][14][15][16] In particular, deep learning methods are highly suitable for de novo drug design. 17 Indeed, some studies 9,18,19 reveal that they have successfully improved drug development by finding more promising compound candidates and raising the success rate of candidate clinical trials.…”

“…Deep learning uses multiple layers of nonlinear transformations that extract and combine information from data in order to distinguish more general trends and patterns from noise, and it also offers the opportunity to acquire features that have not yet been discovered by researchers and to break through the limits of existing knowledge 7–9 . Since deep learning methods can leverage multiple processing layers to automatically extract task‐related features, they have made rapid progress in the fields of image classification, 10 speech recognition, 11 natural language processing, drug discovery, and MD simulations, among others 12–16 . In particular, deep learning methods are highly suitable for de novo drug design 17 .…”