Effective Reaction-Based De Novo Strategy for Kinase Targets: A Case Study on MERTK Inhibitors

Abstract

Reaction-based de novo design is the computational generation of novel molecular structures by linking building blocks using reaction vectors derived from chemistry knowledge. In this work, we first adopted a recurrent neural network (RNN) model to generate three groups of building blocks with different functional groups and then constructed an in silico target-focused combinatorial library based on chemical reaction rules. Mer tyrosine kinase (MERTK) was used as a study case. Combined with a scaffold enrichment analysis, 15 novel MERTK inhibitors covering four scaffolds were achieved. Among them, compound 5a obtained an IC50 value of 53.4 nM against MERTK without any further optimization. The efficiency of hit identification could be significantly improved by shrinking the compound library with the fragment iterative optimization strategy and enriching the dominant scaffold in the hinge region. We hope that this strategy can provide new insights for accelerating the drug discovery process.

The Hitchhiker’s Guide to Deep Learning Driven Generative Chemistry

“…The rational selection from the broader scope of in silico chemical reactions (not just two reactions) and rational iterative nomination of the existing building blocks based on the context of the target should be the focus of ML application rather than generating new building blocks. The authors 71 claimed to perform scaffold analysis to find novel scaffolds applicable to the hinge binding region of MerTK after the docking studies for a designed reaction-based combinatorial library. Five building blocks (Figure 6, 73−77) were picked as scaffolds that were considered to be related to MerTK.…”

“…Hua et al applied RNN to generate functionalized building blocks, which can be utilized in Suzuki and Buchwald–Hartwig reactions in silico to get a virtual combinatorial library of the potential Mer tyrosine kinase (MerTK) inhibitors. 71 Besides this article, there have been reported few research papers utilizing direct accounting for synthetic context during the generation of molecular structures 72 − 74 since the SA of de novo outputs is still an “Achilles heel” of artificial intelligence driven drug design (AIDD). The outputs of generative chemistry are usually triaged before they are submitted for synthesis to focus on only feasible and promising structures.…”

“…The authors 71 claimed to perform scaffold analysis to find novel scaffolds applicable to the hinge binding region of MerTK after the docking studies for a designed reaction-based combinatorial library. Five building blocks ( Figure 6 , 73 – 77 ) were picked as scaffolds that were considered to be related to MerTK.…”

Fake it until you make it? Generative de novo design and virtual screening of synthesizable molecules

“…However, in particular when used in de novo design, generative models based on atomistic representations can be pushed towards generating unreasonable or hard to synthesize molecules, even though to a lesser degree than previous, unconstrained de novo design algorithms, if the exploration is not carefully restricted. [15,30] A promising newer direction are generative models for forward synthesis routes, which similar to the VR approach generate molecules and synthesis routes, leading to much more reasonable structures [56,57,58,45,59]: Bradshaw et al proposed generative models that construct multi-step forward reaction routes where the model learns to pick building blocks and intermediates, which are then submitted to a reaction predictor [56,57]. They demonstrated competitive performance of their algorithm on the Guacamol benchmark [15] compared to less restricted generative models, while maintaining synthesizability.…”

PromptSMILES: Prompting for scaffold decoration and fragment linking in chemical language models

“…In light of alternative and more complex model architectures, CLMs remain either 1 st or 2 nd most performant according to a variety of benchmarks [15][16][17][18] and are the most commonly published deep learning model for de novo molecule generation [19]. Furthermore, they have undergone experimental validation evidencing their ability to generate bioactive molecules de novo in the context of drug design [20][21][22][23][24][25]. Overall CLMs are of increasing utility and importance to augmenting and automating the drug design process.…”