Machine learning in materials science: From explainable predictions to autonomous design

Computational Materials Science, Год журнала: 2021, Номер 193, С. 110360 - 110360

Опубликована: Март 10, 2021

Язык: Английский

---

Benchmarking Machine Learning Models for Polymer Informatics: An Example of Glass Transition Temperature

Journal of Chemical Information and Modeling, Год журнала: 2021, Номер 61(11), С. 5395 - 5413

Опубликована: Окт. 18, 2021

In the field of polymer informatics, utilizing machine learning (ML) techniques to evaluate glass transition temperature Tg and other properties polymers has attracted extensive attention. This data-centric approach is much more efficient practical than laborious experimental measurements when encountered a daunting number structures. Various ML models are demonstrated perform well for prediction. Nevertheless, they trained on different data sets, using structure representations, based feature engineering methods. Thus, critical question arises selecting proper model better handle prediction with generalization ability. To provide fair comparison examine key factors that affect performance, we carry out systematic benchmark study by compiling 79 training them large diverse set. The three major components in setting up an algorithms. terms representation, consider monomer, repeat unit, oligomer longer chain structure. Based feature, representation calculated, including Morgan fingerprinting or without substructure frequency, RDKit descriptors, molecular embedding, graph, etc. Afterward, obtained input algorithms, such as deep neural networks, convolutional random forest, support vector machine, LASSO regression, Gaussian process regression. We performance these holdout test set extra unlabeled from high-throughput dynamics simulation. model's ability especially focused, sensitivity topology weight also taken into consideration. provides not only guideline task but useful reference informatics tasks.

Язык: Английский

---

Machine Learning on a Robotic Platform for the Design of Polymer–Protein Hybrids

Advanced Materials, Год журнала: 2022, Номер 34(30)

Опубликована: Май 20, 2022

Abstract Polymer–protein hybrids are intriguing materials that can bolster protein stability in non‐native environments, thereby enhancing their utility diverse medicinal, commercial, and industrial applications. One stabilization strategy involves designing synthetic random copolymers with compositions attuned to the surface, but rational design is complicated by vast chemical composition space. Here, a reported protein‐stabilizing based on active machine learning, facilitated automated material synthesis characterization platforms. The versatility robustness of approach demonstrated successful identification preserve, or even enhance, activity three chemically distinct enzymes following exposure thermal denaturing conditions. Although systematic screening results mixed success, learning appropriately identifies unique effective copolymer chemistries for each enzyme. Overall, this work broadens capabilities fit‐for‐purpose promote otherwise manipulate activity, extensions toward robust polymer–protein hybrid materials.

Язык: Английский

---

Bioactive Synthetic Polymers

Advanced Materials, Год журнала: 2021, Номер 34(2)

Опубликована: Окт. 5, 2021

Abstract Synthetic polymers are omnipresent in society as textiles and packaging materials, construction medicine, among many other important applications. Alternatively, natural play a crucial role sustaining life allowing organisms to adapt their environments by performing key biological functions such molecular recognition transmission of genetic information. In general, the synthetic polymer worlds completely separated due inability for perform specific functions; some cases, cause uncontrolled unwanted responses. However, owing advancement polymerization techniques recent years, new have emerged that provide targeted peptides, or present antiviral, anticancer, antimicrobial activities. this review, emergence generation bioactive bioapplications summarized. Finally, future opportunities area discussed.

Язык: Английский

---

Machine-Learning-Guided Discovery of ¹⁹F MRI Agents Enabled by Automated Copolymer Synthesis

Journal of the American Chemical Society, Год журнала: 2021, Номер 143(42), С. 17677 - 17689

Опубликована: Окт. 12, 2021

Modern polymer science suffers from the curse of multidimensionality. The large chemical space imposed by including combinations monomers into a statistical copolymer overwhelms synthesis and characterization technology limits ability to systematically study structure–property relationships. To tackle this challenge in context 19F magnetic resonance imaging (MRI) agents, we pursued computer-guided materials discovery approach that combines synergistic innovations automated flow machine learning (ML) method development. A software-controlled, continuous platform was developed enable iterative experimental–computational cycles resulted 397 unique compositions within six-variable compositional space. nonintuitive design criteria identified ML, which were accomplished exploring <0.9% overall space, lead identification >10 outperformed state-of-the-art materials.

Язык: Английский

---

polyBERT: a chemical language model to enable fully machine-driven ultrafast polymer informatics

Nature Communications, Год журнала: 2023, Номер 14(1)

Опубликована: Июль 11, 2023

Polymers are a vital part of everyday life. Their chemical universe is so large that it presents unprecedented opportunities as well significant challenges to identify suitable application-specific candidates. We present complete end-to-end machine-driven polymer informatics pipeline can search this space for candidates at speed and accuracy. This includes fingerprinting capability called polyBERT (inspired by Natural Language Processing concepts), multitask learning approach maps the fingerprints host properties. linguist treats structure polymers language. The outstrips best presently available concepts property prediction based on handcrafted fingerprint schemes in two orders magnitude while preserving accuracy, thus making strong candidate deployment scalable architectures including cloud infrastructures.

Язык: Английский

---

A graph representation of molecular ensembles for polymer property prediction

Chemical Science, Год журнала: 2022, Номер 13(35), С. 10486 - 10498

Опубликована: Янв. 1, 2022

Synthetic polymers are versatile and widely used materials. Similar to small organic molecules, a large chemical space of such materials is hypothetically accessible. Computational property prediction virtual screening can accelerate polymer design by prioritizing candidates expected have favorable properties. However, in contrast often not well-defined single structures but an ensemble similar which poses unique challenges traditional representations machine learning approaches. Here, we introduce graph representation molecular ensembles associated neural network architecture that tailored prediction. We demonstrate this approach captures critical features polymeric materials, like chain architecture, monomer stoichiometry, degree polymerization, achieves superior accuracy off-the-shelf cheminformatics methodologies. While doing so, built dataset simulated electron affinity ionization potential values for >40k with varying composition, may be the development other The models presented work pave path toward new classes algorithms informatics and, more broadly, framework modeling ensembles.

Язык: Английский

---

TransPolymer: a Transformer-based language model for polymer property predictions

npj Computational Materials, Год журнала: 2023, Номер 9(1)

Опубликована: Апрель 22, 2023

Abstract Accurate and efficient prediction of polymer properties is great significance in design. Conventionally, expensive time-consuming experiments or simulations are required to evaluate functions. Recently, Transformer models, equipped with self-attention mechanisms, have exhibited superior performance natural language processing. However, such methods not been investigated sciences. Herein, we report TransPolymer, a Transformer-based model for property prediction. Our proposed tokenizer chemical awareness enables learning representations from sequences. Rigorous on ten benchmarks demonstrate the TransPolymer. Moreover, show that TransPolymer benefits pretraining large unlabeled dataset via Masked Language Modeling. Experimental results further manifest important role modeling We highlight this as promising computational tool promoting rational design understanding structure-property relationships data science view.

Язык: Английский

---

Emerging Trends in Machine Learning: A Polymer Perspective

ACS Polymers Au, Год журнала: 2023, Номер 3(3), С. 239 - 258

Опубликована: Янв. 18, 2023

In the last five years, there has been tremendous growth in machine learning and artificial intelligence as applied to polymer science. Here, we highlight unique challenges presented by polymers how field is addressing them. We focus on emerging trends with an emphasis topics that have received less attention review literature. Finally, provide outlook for field, outline important areas science discuss advances from greater material community.

Язык: Английский

---

Scope of machine learning in materials research—A review

Applied Surface Science Advances, Год журнала: 2023, Номер 18, С. 100523 - 100523

Опубликована: Ноя. 28, 2023

This comprehensive review investigates the multifaceted applications of machine learning in materials research across six key dimensions, redefining field's boundaries. It explains various knowledge acquisition mechanisms starting with supervised, unsupervised, reinforcement, and deep techniques. These techniques are transformative tools for transforming unactionable data into insightful actions. Moving on to synthesis, emphasizes profound influence learning, as demonstrated by predictive models that speed up material selection, structure-property relationships reveal crucial connections, data-driven discovery fosters innovation. Machine reshapes our comprehension manipulation accelerating enabling tailored design through property prediction relationships. extends its image processing, improving object detection, classification, segmentation precision methods like generation, revolutionizing potential processing research. The most recent developments show how can have a impact at atomic level precise intricate extraction, representing significant advancements understanding highlights has revolutionize discovery, performance, stimulating does so while acknowledging obstacles poor quality complicated algorithms. offers wide range exciting prospects scientific investigation technological advancement, positioning it powerful force influencing future

Язык: Английский

---