Machine learning-aided generative molecular design

Du, Yuanqi; Jamasb, Arian R.; Guo, Jeff; Fu, Tianfan; Harris, Charles; Wang, Yingheng; Duan, Chenru; Liò, Pietro; Schwaller, Philippe; Blundell, Tom L.

doi:10.1038/s42256-024-00843-5

Cited by 3 publications

(9 citation statements)

References 117 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Atomwise published an average hit rate of 8.8% across 22 targets; however, this encompasses hits at concentrations up to 250 μM 6 . Insilico Medicine has reported an average of 28% across 2 targets, though this includes hits at concentrations up to 25 μM and lead optimization-phase hits 7,8,9 . These AI approaches not only have limited hit rates but also report non-standardized concentrations to define "hits", likely leading to inflated hit rates that permit concentrations that are too high to be therapeutically ), which encodes structural features into a binary vector.…”

Section: Challenges With Ai Drug Discoverymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…viable 5,6,7,8,9 . Furthermore, few of the compounds discovered as hits demonstrate significant chemical novelty, highlighting the inability of these models to predict properties of compounds beyond the chemical space they have seen 10,11 .…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery

Umansky,

Woods,

Russell

et al. 2024

Preprint

View full text Add to dashboard Cite

In the evolving landscape of drug discovery, traditional methods face significant challenges due to low efficiency and high resource demands. This study introduces our GALILEO AI drug discovery platform and its foundational ChemPrint model, designed to improve the efficiency of drug discovery. Addressing the challenges of low hit rates and the difficulty in exploring novel chemical spaces, this platform adopts adaptive molecular embeddings and stringent model training environments to enhance predictive capabilities and navigate uncharted molecular territories. We validate this approach through a case study targeting the AXL and BRD4 oncology targets, where ChemPrint achieved a 45.5% in vitro hit rate and identified 20 novel acting compounds. These compounds exhibited substantial chemical novelty, with an average Tanimoto similarity score of 0.32 to their training set, significantly diverging from known compounds and demonstrating ChemPrint's capability to extrapolate, surpassing traditional benchmarks and industry standards. This underscores the platform's ability to bridge the gap between the potential of AI and its practical application in therapeutic discovery.

show abstract

Section: Challenges With Ai Drug Discoverymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery

Umansky,

Woods,

Russell

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…In light of this expected error between optimized proxy and true objective, it is important to propose diverse solutions to increase the chance of identifying a molecule of interest with all desirable properties for further development - another contrast to classical RL where a single successful solution is often sufficient. Despite these challenges, RL has shown promising preliminary results with successful application to drug design across a number of different molecular representations and model architectures …”

Section: Introductionmentioning

confidence: 99%

ACEGEN: Reinforcement Learning of Generative Chemical Agents for Drug Discovery

Bou,

Thomas,

Dittert

et al. 2024

J. Chem. Inf. Model.

View full text Add to dashboard Cite

In recent years, reinforcement learning (RL) has emerged as a valuable tool in drug design, offering the potential to propose and optimize molecules with desired properties. However, striking a balance between capabilities, flexibility, reliability, and efficiency remains challenging due to the complexity of advanced RL algorithms and the significant reliance on specialized code. In this work, we introduce ACEGEN, a comprehensive and streamlined toolkit tailored for generative drug design, built using TorchRL, a modern RL library that offers thoroughly tested reusable components. We validate ACEGEN by benchmarking against other published generative modeling algorithms and show comparable or improved performance. We also show examples of ACEGEN applied in multiple drug discovery case studies. ACEGEN is accessible at https://github.com/acellera/acegen-open and available for use under the MIT license.

show abstract

“…Instead, there are complex chemical rules that dictate, for instance, each atom's number of eligible bonds, possible charges or bond angles and these rules are difficult to represent. While several molecular representations exist [15], many of them are either discrete, not unique, or very sparse, whereas a continuous numeric representation is the preferred input for many deep learning models. Latent diffusion models outsource the task of learning a mapping from the molecular representation to a continuous latent space to a VAE, while the training of the diffusion model itself is focussed on modelling its latent distribution.…”

mentioning

confidence: 99%

DrugDiff - small molecule diffusion model with flexible guidance towards molecular properties

Oestreich,

Merdivan,

Lee

et al. 2024

Preprint

View full text Add to dashboard Cite

With the cost/yield-ratio of drug development becoming increasingly unfavourable, recent work has explored machine learning to accelerate early stages of the development process. Given the current success of deep generative models across domains, we here investigated their application to the property-based proposal of new small molecules for drug development. Specifically, we trained a latent diffusion model - DrugDiff - paired with predictor guidance to generate novel compounds with a variety of desired molecular properties. The architecture was designed to be highly flexible and easily adaptable to future scenarios. Our experiments showed successful generation of unique, diverse and novel small molecules with targeted properties. The code is available at https://github.com/MarieOestreich/DrugDiff.

show abstract

Machine learning-aided generative molecular design

Cited by 3 publications

References 117 publications

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery

ACEGEN: Reinforcement Learning of Generative Chemical Agents for Drug Discovery

DrugDiff - small molecule diffusion model with flexible guidance towards molecular properties

Contact Info

Product

Resources

About