Deep learning for molecular design—a review of the state of the art

Elton, Daniel C.; Boukouvalas, Zois; Fuge, Mark; Chung, Peter

doi:10.1039/c9me00039a

Cited by 507 publications

(508 citation statements)

References 121 publications

Supporting

Mentioning

497

Contrasting

Unclassified

Order By: Relevance

“…8 In the past decade, following on the advent of Variational Auto-Encoders (VAEs) 9 and Generative Adversarial Networks (GANs), 10 there has been a flood of new deep learning (DL) methods for this task. 11 Many of these methods learn a mapping from a continuous lower-dimensional real number space to a discrete chemical space. Jointly trained with a structure-property regression, one can obtain novel chemical structures conditioned on desired properties.…”

Section: Introductionmentioning

confidence: 99%

The Synthesizability of Molecules Proposed by Generative Models

Gao

Coley

2020

J. Chem. Inf. Model.

212

248

View full text Add to dashboard Cite

The discovery of functional molecules is an expensive and time-consuming process, exemplified by the rising costs of small molecule therapeutic discovery. One class of techniques of growing interest for early-stage drug discovery is de novo molecular generation and optimization, catalyzed by the development of new deep learning approaches. 1 These techniques can suggest novel molecular structures intended to maximize a multi-objective function, e.g., suitability as a therapeutic against a particular target, 2 without relying on brute-force exploration of a chemical space. 3 However, the utility of these approaches is stymied by ignorance of synthesizability. To highlight the severity of this issue, we use a data-driven computer-aided synthesis planning program 4 to quantify how often molecules proposed by state-of-the-art generative models cannot be readily synthesized. Our analysis demonstrates that there are several tasks for which these models generate unrealistic molecular structures despite performing well on popular quantitative benchmarks. Synthetic complexity heuristics can successfully bias generation toward synthetically-tractable chemical space, although doing so 1 arXiv:2002.07007v1 [q-bio.QM] 17 Feb 2020 necessarily detracts from the primary objective. This analysis suggests that to improve the utility of these models in real discovery workflows, new algorithm development is warranted.

show abstract

Section: Introductionmentioning

confidence: 99%

The Synthesizability of Molecules Proposed by Generative Models

Gao

Coley

2020

J. Chem. Inf. Model.

212

248

View full text Add to dashboard Cite

show abstract

“…These methods include Variational AutoEncoders (VAEs), Recurrent Neural Network (RNN), Generative Adversarial Networks (GANs) and reinforcement learning (RL) [ 11 ]. Another alternative to build generative molecules models without SMILES is based on molecular graph representation [11]. Contrary to earlier reports [11], we demonstrate herein that text learning on SMILES is highly efficient to explore the training space with a high degree of novelty.…”

Section: Introductionmentioning

confidence: 88%

“…Since 2016, SMILES-based machine-learned methods are used to produce de novo molecules. These methods include Variational AutoEncoders (VAEs), Recurrent Neural Network (RNN), Generative Adversarial Networks (GANs) and reinforcement learning (RL) [ 11 ]. Another alternative to build generative molecules models without SMILES is based on molecular graph representation [11].…”

Section: Introductionmentioning

confidence: 99%

GEN: Highly Efficient SMILES Explorer Using Autodidactic Generative Examination Networks

Deursen

Ertl

Tetko

et al. 2019

Preprint

View full text Add to dashboard Cite

Recurrent neural networks have been widely used to generate millions of de novo molecules in a known chemical space. These deep generative models are typically setup with LSTM or GRU units and trained with canonical SMILES. In this study, we introduce a new robust architecture, Generative Examination Network GEN, based on bidirectional RNNs with concatenated sub-models to learn and generate molecular SMILES within a trained target space. GENs autonomously learn the target space in a few epochs while being subjected to an independent online examination to measure the quality of the generated set. Here we have used online statistical quality control (SQC) on the percentage of valid molecular SMILES as examination measure to select the earliest available stable model weights. Very high levels of valid SMILES (95-98%) can be generated using multiple parallel encoding layers in combination with SMILES augmentation using unrestricted SMILES randomization. Our architecture combines an excellent novelty rate (85-90%) while generating SMILES with strong conservation of the property space (95-99%). Our flexible examination mechanism is open to other quality criteria.

show abstract

“…In total, 4,922 unique valid structures were automatically generated and all matched the defined rules by using ADQN-FBDD without any pre-training as many other methods need. 35,40,41,54,55 Next, All the molecules with high deep reinforcement learning scores (DRL score: R(S)>0.6) were kept (47 molecules). Then, these 47 unique molecules were prepared to generate at least 1 conformation with the local energy minimization using the OPLS-2005 force field by the "ligand prepare" module of Schrödinger 2015 software.…”

Section: Molecular Generation and Selectionmentioning

confidence: 99%

AI-aided design of novel targeted covalent inhibitors against SARS-CoV-2

Tang

Liu

et al. 2020

Preprint

View full text Add to dashboard Cite

The focused drug repurposing of known approved drugs (such as lopinavir/ritonavir) has been reported failed for curing SARS-CoV-2 infected patients. It is urgent to generate new chemical entities against this virus. As a key enzyme in the life-cycle of coronavirus, the 3C-like main protease (3CL pro or M pro ) is the most attractive for antiviral drug design. Based on a recently solved structure (PDB ID: 6LU7), we developed a novel advanced deep Q-learning network with the fragment-based drug design (ADQN-FBDD) for generating potential lead compounds targeting SARS-CoV-2 3CL pro . We obtained a series of derivatives from those lead compounds by our structure-based optimization policy (SBOP). All the 47 lead compounds directly from our AI-model and related derivatives based on SBOP are accessible in our molecular library at https://github.com/tbwxmu/2019-nCov. These compounds can be used as potential candidates for researchers in their development of drugs against SARS-CoV-2.author/funder. All rights reserved. No reuse allowed without permission. : bioRxiv preprint disease (COVID-19) worldwide. 1 As of March 2, 2020, more than 90,000 people have been infected by SARS-CoV-2 and more than 3000 people have been reported dead according to Johns Hopkins Coronavirus map tracker. 2 The numbers of infection and death are still increasing. To face the considerable threat of SARS-CoV-2, it is urgent to develop new inhibitors or drugs against this deadly virus. Unfortunately, since the outbreak of severe acute respiratory syndrome (SARS) eighteen years ago, there has been no approved treatment against the SARS coronavirus (SARS-CoV), 3 which is similar to SARS-CoV-2. Repurposing potential drugs such as lopinavir and ritonavir also failed to SARS-CoV-2 injected patients. 4 Structure-based antiviral drug design with a new artificial intelligence algorithm may represent a more helpful approach to get the SARS-CoV-2 targeted inhibitors or drugs. Thanks to the prompt efforts of many researchers, we have several pieces of important information about this vital virus genome and protein structures. We now know that the non-structural protein 5 (Nsp5) is the main protease (M pro ) of SARS-CoV-2 and it is a cysteine protease, which also been called "3C-like protease" (3CL pro ). Moreover, we know that the 3D structure of 3CL pro is very similar to SARS-CoV with a sequence identity of >96% and 3D structure superposition RMSDCα of 0.44 Å as shown in Figures S1 and S2. 3CL pro has been reported as an attractive target for developing anti-coronaviral drugs: 1) this protease is highly conserved in both sequences and 3D structures; 5 2) 3CL pro is a key enzyme for related virus (including SARS and SARS-CoV-2) replication; 3) it only exists in the virus, not in humans. Developing specific antiviral drugs targeting 3CL pro of the specific virus has shown significant success; for example, both approved drugs lopinavir and ritonavir can completely occupy the substrate-binding site of 3CL pro to break down the replication of human immunodeficien...

show abstract

Deep learning for molecular design—a review of the state of the art

Cited by 507 publications

References 121 publications

The Synthesizability of Molecules Proposed by Generative Models

The Synthesizability of Molecules Proposed by Generative Models

GEN: Highly Efficient SMILES Explorer Using Autodidactic Generative Examination Networks

AI-aided design of novel targeted covalent inhibitors against SARS-CoV-2

Contact Info

Product

Resources

About