Applications of Deep Learning in Molecule Generation and Molecular Property Prediction

Accounts of Chemical Research, Год журнала: 2020, Номер 54(2), С. 263 - 270

Опубликована: Дек. 28, 2020

ConspectusRecent advances in computer hardware and software have led to a revolution deep neural networks that has impacted fields ranging from language translation vision. Deep learning also number of areas drug discovery, including the analysis cellular images design novel routes for synthesis organic molecules. While work these been impactful, complete review applications discovery would be beyond scope single Account. In this Account, we will focus on two key where molecular design: prediction properties de novo generation suggestions new molecules.One most significant development quantitative structure–activity relationships (QSARs) come application methods biological activity physical molecules programs. Rather than employing expert-derived chemical features typically used build predictive models, researchers are now using develop representations. These representations, coupled with ability uncover complex, nonlinear relationships, state-of-the-art performance. changed way many approach QSARs, it is not panacea. As any other machine task, models dependent quality, quantity, relevance available data. Seemingly fundamental issues, such as optimal creating training set, still open questions field. Another critical area subject multiple research efforts assessing confidence model.Deep contributed renaissance molecule generation. relying manually defined heuristics, learn generate based sets existing Techniques were originally developed image adapted described above being specific predicted profiles. generative algorithms appear promising, there only few reports testing designs proposed by models. The evaluation diversity, ultimate value produced an question. field benchmarks, yet agree how one should ultimately assess "invented" algorithm.

Язык: Английский

---

Generative models for molecular discovery: Recent advances and challenges

Wiley Interdisciplinary Reviews Computational Molecular Science, Год журнала: 2022, Номер 12(5)

Опубликована: Март 5, 2022

Abstract Development of new products often relies on the discovery novel molecules. While conventional molecular design involves using human expertise to propose, synthesize, and test molecules, this process can be cost time intensive, limiting number molecules that reasonably tested. Generative modeling provides an alternative approach by reformulating as inverse problem. Here, we review recent advances in state‐of‐the‐art generative discusses considerations for integrating these models into real campaigns. We first model choices required develop train a including common 1D, 2D, 3D representations typical neural network architectures. then describe different problem statements applications explore benchmarks used evaluate based those statements. Finally, discuss important factors play role experimental workflows. Our aim is will equip reader with information context necessary utilize within their domain. This article categorized under: Data Science > Artificial Intelligence/Machine Learning

Язык: Английский

---

Machine Learning Methods for Small Data Challenges in Molecular Science

Chemical Reviews, Год журнала: 2023, Номер 123(13), С. 8736 - 8780

Опубликована: Июнь 29, 2023

Small data are often used in scientific and engineering research due to the presence of various constraints, such as time, cost, ethics, privacy, security, technical limitations acquisition. However, big have been focus for past decade, small their challenges received little attention, even though they technically more severe machine learning (ML) deep (DL) studies. Overall, challenge is compounded by issues, diversity, imputation, noise, imbalance, high-dimensionality. Fortunately, current era characterized technological breakthroughs ML, DL, artificial intelligence (AI), which enable data-driven discovery, many advanced ML DL technologies developed inadvertently provided solutions problems. As a result, significant progress has made decade. In this review, we summarize analyze several emerging potential molecular science, including chemical biological sciences. We review both basic algorithms, linear regression, logistic regression (LR),

Язык: Английский

---

Machine Learning in Chemical Engineering: A Perspective

Chemie Ingenieur Technik, Год журнала: 2021, Номер 93(12), С. 2029 - 2039

Опубликована: Окт. 22, 2021

Abstract The transformation of the chemical industry to renewable energy and feedstock supply requires new paradigms for design flexible plants, (bio‐)catalysts, functional materials. Recent breakthroughs in machine learning (ML) provide unique opportunities, but only joint interdisciplinary research between ML engineering (CE) communities will unfold full potential. We identify six challenges that open methods CE formulate types problems ML: (1) optimal decision making, (2) introducing enforcing physics ML, (3) information knowledge representation, (4) heterogeneity data, (5) safety trust applications, (6) creativity. Under umbrella these challenges, we discuss perspectives future enable CE.

Язык: Английский

---

Evidential Deep Learning for Guided Molecular Property Prediction and Discovery

ACS Central Science, Год журнала: 2021, Номер 7(8), С. 1356 - 1367

Опубликована: Июль 27, 2021

While neural networks achieve state-of-the-art performance for many molecular modeling and structure-property prediction tasks, these models can struggle with generalization to out-of-domain examples, exhibit poor sample efficiency, produce uncalibrated predictions. In this paper, we leverage advances in evidential deep learning demonstrate a new approach uncertainty quantification network-based at no additional computational cost. We develop both 2D message passing 3D atomistic apply across range of different tasks. that uncertainties enable (1) calibrated predictions where correlates error, (2) sample-efficient training through uncertainty-guided active learning, (3) improved experimental validation rates retrospective virtual screening campaign. Our results suggest provide an efficient means useful property prediction, discovery, design tasks the chemical physical sciences.

Язык: Английский

---

Machine learning enables interpretable discovery of innovative polymers for gas separation membranes

Science Advances, Год журнала: 2022, Номер 8(29)

Опубликована: Июль 20, 2022

Polymer membranes perform innumerable separations with far-reaching environmental implications. Despite decades of research, design new membrane materials remains a largely Edisonian process. To address this shortcoming, we demonstrate generalizable, accurate machine learning (ML) implementation for the discovery innovative polymers ideal performance. Specifically, multitask ML models are trained on experimental data to link polymer chemistry gas permeabilities He, H2, O2, N2, CO2, and CH4. We interpret extract valuable insights into contributions different chemical moieties permeability selectivity. then screen over 9 million hypothetical identify thousands that lie well above current performance upper bounds, including hundreds never-before-seen ultrapermeable O2 CO2 greater than 104 105 Barrers, respectively. High-fidelity molecular dynamics simulations confirm ML-predicted promising candidates, which suggests many can be translated reality.

Язык: Английский

---

Deep learning methods for molecular representation and property prediction

Drug Discovery Today, Год журнала: 2022, Номер 27(12), С. 103373 - 103373

Опубликована: Сен. 25, 2022

Язык: Английский

---

Closed-loop optimization of general reaction conditions for heteroaryl Suzuki-Miyaura coupling

Science, Год журнала: 2022, Номер 378(6618), С. 399 - 405

Опубликована: Окт. 27, 2022

General conditions for organic reactions are important but rare, and efforts to identify them usually consider only narrow regions of chemical space. Discovering more general reaction requires considering vast space derived from a large matrix substrates crossed with high-dimensional conditions, rendering exhaustive experimentation impractical. Here, we report simple closed-loop workflow that leverages data-guided down-selection, uncertainty-minimizing machine learning, robotic discover conditions. Application the challenging consequential problem heteroaryl Suzuki-Miyaura cross-coupling identified double average yield relative widely used benchmark was previously developed using traditional approaches. This study provides practical road map solving multidimensional optimization problems search spaces.

Язык: Английский

---

A Novel Approach to Uncertainty Quantification in Groundwater Table Modeling by Automated Predictive Deep Learning

Natural Resources Research, Год журнала: 2022, Номер 31(3), С. 1351 - 1373

Опубликована: Апрель 12, 2022

Abstract Uncertainty quantification ( UQ ) is an important benchmark to assess the performance of artificial intelligence AI and particularly deep learning ensembled-based models. However, ability for using current -based methods not only limited in terms computational resources but it also requires changes topology optimization processes, as well multiple performances monitor model instabilities. From both geo-engineering societal perspectives, a predictive groundwater table GWT presents challenge, where lack limits validity findings may undermine science-based decisions. To overcome address these limitations, novel ensemble, automated random deactivating connective weights approach ARDCW ), presented applied retrieved geographical locations data from project Stockholm, Sweden. In this approach, was achieved via combination several derived ensembles fixed optimum subjected randomly switched off weights, which allow predictability with one forward pass. The process developed programmed provide trackable specific task access wide variety different internal characteristics libraries. A comparison Monte Carlo dropout quantile regression computer vision control metrics showed significant progress . This does require can be already trained topologies way that outperforms other

Язык: Английский

---

Data‐Driven Materials Innovation and Applications

Advanced Materials, Год журнала: 2022, Номер 34(36)

Опубликована: Апрель 22, 2022

Abstract Owing to the rapid developments improve accuracy and efficiency of both experimental computational investigative methodologies, massive amounts data generated have led field materials science into fourth paradigm data‐driven scientific research. This transition requires development authoritative up‐to‐date frameworks for approaches material innovation. A critical discussion on current advances in discovery with a focus frameworks, machine‐learning algorithms, material‐specific databases, descriptors, targeted applications inorganic is presented. Frameworks rationalizing innovation are described, review essential subdisciplines presented, including: i) advanced data‐intensive strategies algorithms; ii) databases related tools platforms generation management; iii) commonly used molecular descriptors processes. Furthermore, an in‐depth broad innovation, such as energy conversion storage, environmental decontamination, flexible electronics, optoelectronics, superconductors, metallic glasses, magnetic materials, provided. Finally, how these (with insights synergy science, tools, mathematics) support paradigms outlined, opportunities challenges highlighted.

Язык: Английский

---