Link-INVENT: generative linker design with reinforcement learning

Guo, Jeff; Knuth, Franziska; Margreitter, Christian; Janet, Jon Paul; Papadopoulos, Kostas; Engkvist, Ola; Patronov, Atanas

doi:10.1039/d2dd00115b

Cited by 23 publications

(27 citation statements)

References 66 publications

(297 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The resulting samples, therefore, resemble both valid-looking molecules and the reference set of coordinates. The ability for reference atoms of fragments to move during generation distinguishes SILVR from the standard linker design. , …”

Section: Methodsmentioning

confidence: 99%

“…The ability for reference atoms of fragments to move during generation distinguishes SILVR from the standard linker design. 24,27 Reference Dataset: COVID Moonshot. Reference molecules were selected from the original 23 noncovalent hits of the SARS-CoV-2 main protease (Mpro) identified by the XChem fragment screen 40 as part of the COVID Moonshot Project.…”

Section: ■ Methodsmentioning

confidence: 99%

“…An application of generative models is, therefore, in the interpretation of such fragment data for the automated design of high-affinity binders. Goa et al have introduced a way to generate linkers for fragments using reinforcement learning strategies, 24 while Imrie et al 25 have used variational autoencoders on this task without the use of protein information. More recently, Huang et al 26 and Igashov et al 27 have tackled this challenge using equivariant variational autoencoder and diffusion-based models, respectively.…”

Section: ■ Introductionmentioning

confidence: 99%

See 2 more Smart Citations

SILVR: Guided Diffusion for Molecule Generation

Runcie,

Mey

2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Computationally generating new synthetically accessible compounds with high affinity and low toxicity is a great challenge in drug design. Machine learning models beyond conventional pharmacophoric methods have shown promise in the generation of novel small-molecule compounds but require significant tuning for a specific protein target. Here, we introduce a method called selective iterative latent variable refinement (SILVR) for conditioning an existing diffusion-based equivariant generative model without retraining. The model allows the generation of new molecules that fit into a binding site of a protein based on fragment hits. We use the SARS-CoV-2 main protease fragments from Diamond XChem that form part of the COVID Moonshot project as a reference dataset for conditioning the molecule generation. The SILVR rate controls the extent of conditioning, and we show that moderate SILVR rates make it possible to generate new molecules of similar shape to the original fragments, meaning that the new molecules fit the binding site without knowledge of the protein. We can also merge up to 3 fragments into a new molecule without affecting the quality of molecules generated by the underlying generative model. Our method is generalizable to any protein target with known fragments and any diffusion-based model for molecule generation.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: ■ Methodsmentioning

confidence: 99%

Section: ■ Introductionmentioning

confidence: 99%

See 1 more Smart Citation

SILVR: Guided Diffusion for Molecule Generation

Runcie,

Mey

2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

show abstract

“…DR-linker [32], Link-Invent [33], DiffLinker [34] and De-linker [35] algorithms were developed for linker generation; GraphGMVAE [36], and Deep-Hops [37] methods were proposed for scaffold hopping; Scaffold Decorator [38], SAMOA [39] and Lib-Invent [40] were dedicated for scaffold decoration. So far to best of our knowledge, to gain precise structure control in these various scenarios would need using generative model with different architectures and there is no single model can address all these tasks.…”

Section: Structure Generation With Topology Constrainsmentioning

confidence: 99%

Tree-Invent: A novel molecular generative model constrained with topological tree

Xu¹,

Chen²

2023

Preprint

View full text Add to dashboard Cite

De novo molecular design plays an important role in drug discovery. Here a novel generative model, Tree-Invent, was proposed to integrate topological constraints in the generation of molecular graph. In this model, a molecular graph is represented as a topological tree in which ring system, non-ring atom and chemical bond are regarded as ring node, single node and edge respectively. The molecule generation is driven by three independent sub-models for carrying out operations of node addition, ring generation and node connection. One unique feature of the generative model is that topological tree structure can be specified as constraint for structure generation, which provides more precise control on structure generation. Additionally, the quality of ring structure is also improved by reducing the occurrence of macrocycle during structure generation. Combining with reinforcement learning, Tree-Invent model could efficiently explore targeted chemical space. Moreover, Tree-Invent model is flexible enough to be used in versatile molecule design settings such as scaffold decoration, scaffold hopping and linker generation.

show abstract

“…Further, a gated graph neural network (GGNN) outperforms molecular graph generation in deep generative models 28,29 and demonstrates the practical structure formation in drug design. 20,30 Many approaches use two-dimensional (2D) SMILES-based chemical graphs embedded in lowdimensional space to generate new molecules by perturbing the hidden values of the sampled atoms.…”

Section: ■ Introductionmentioning

confidence: 99%

Fragment Linker Prediction Using the Deep Encoder-Decoder Network for PROTACs Drug Design

Kao¹,

Lin

Chou³

et al. 2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

A drug discovery and development pipeline is a prolonged and complex process that remains challenging for both computational methods and medicinal chemists and has not been able to be resolved using computational methods. Deep learning has been utilized in various fields and achieved tremendous success in designing novel molecules in the pharmaceutical industry. Herein, we use state-of-the-art techniques to propose a deep neural network, AIMLinker, to rapidly design and generate meaningful drug-like proteolysis targeting chimeras (PROTACs) analogs. The model extracts the structural information from the input fragments and generates linkers to incorporate them. We integrate filters in the model to exclude nondruggable structures guided via protein–protein complexes while retaining molecules with potent chemical properties. The novel PROTACs subsequently pass through molecular docking, taking root-mean-square deviation (RMSD), relative Gibbs free energy (ΔΔG binding ), molecular dynamics (MD) simulation, and free energy perturbation (FEP) calculations as the measurement criteria for testing the robustness and feasibility of the model. The generated novel PROTACs molecules possess similar structural information with superior binding affinity to the binding pockets compared to the existing CRBN-dBET6-BRD4 ternary complexes. We demonstrate the effectiveness of the methodology of leveraging AIMLinker to design novel compounds for PROTACs molecules exhibiting better chemical properties compared to the dBET6 crystal pose.

show abstract

Link-INVENT: generative linker design with reinforcement learning

Abstract: In this work, we present Link-INVENT as an extension to the existing de novo molecular design platform REINVENT. We provide illustrative examples on how Link-INVENT can be applied on fragment...

Cited by 23 publications

References 66 publications

SILVR: Guided Diffusion for Molecule Generation

SILVR: Guided Diffusion for Molecule Generation

Tree-Invent: A novel molecular generative model constrained with topological tree

Fragment Linker Prediction Using the Deep Encoder-Decoder Network for PROTACs Drug Design

Contact Info

Product

Resources

About