Molecular Machine Learning: The Future of Synthetic Chemistry?

Pflüger, Philipp Miro; Glorius, Frank

doi:10.1002/anie.202008366

Cited by 58 publications

(60 citation statements)

References 47 publications

(53 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…266 Püger and Glorius mention that in order for us to understand what machines learn "XAI must nd its way into chemistry", which requires adequate understanding of both the algorithms and the chemistry. 267 Indeed, extremely complex algorithms that are only understood by a specialised computer scientist and input data that are only understood by a theoretical chemist are the bottleneck for the progress of this eld. Such systems have started being implemented in several chemistry-related studies, 265,268,269 but they are still not popular in IL research.…”

Section: Future Aspectsmentioning

confidence: 99%

A review on machine learning algorithms for the ionic liquid chemical space

et al. 2021

View full text Add to dashboard Cite

show abstract

Section: Future Aspectsmentioning

confidence: 99%

A review on machine learning algorithms for the ionic liquid chemical space

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Originally proposed by Vaswani and colleagues [11], transformers have come to dominate the list of preferred methods, especially those used with strings such as those involved in natural language processing [106,[160][161][162][163]. Since chemical structures can be encoded as strings such as SMILES [164], it is clear that transformers might be used with success to attach problems involving small molecules, and they have indeed been so exploited (e.g., [10,12,104,[165][166][167][168]). In the present work, we have adopted and refined the transformer architecture.…”

Section: Discussionmentioning

confidence: 99%

FragNet, a Contrastive Learning-Based Transformer Model for Clustering, Interpreting, Visualizing, and Navigating Chemical Space

Shrivastava

Kell

2021

Molecules

View full text Add to dashboard Cite

The question of molecular similarity is core in cheminformatics and is usually assessed via a pairwise comparison based on vectors of properties or molecular fingerprints. We recently exploited variational autoencoders to embed 6M molecules in a chemical space, such that their (Euclidean) distance within the latent space so formed could be assessed within the framework of the entire molecular set. However, the standard objective function used did not seek to manipulate the latent space so as to cluster the molecules based on any perceived similarity. Using a set of some 160,000 molecules of biological relevance, we here bring together three modern elements of deep learning to create a novel and disentangled latent space, viz transformers, contrastive learning, and an embedded autoencoder. The effective dimensionality of the latent space was varied such that clear separation of individual types of molecules could be observed within individual dimensions of the latent space. The capacity of the network was such that many dimensions were not populated at all. As before, we assessed the utility of the representation by comparing clozapine with its near neighbors, and we also did the same for various antibiotics related to flucloxacillin. Transformers, especially when as here coupled with contrastive learning, effectively provide one-shot learning and lead to a successful and disentangled representation of molecular latent spaces that at once uses the entire training set in their construction while allowing “similar” molecules to cluster together in an effective and interpretable way.

show abstract

“…To overcome these obstacles, chemists have started to combine ML with real-time chemistry. Indeed, the ability of ML to recognize complex patterns in data bears the potential to modernize the way chemical challenges are approached [78][79][80]. ML is a subset of artificial intelligence, defined as the field of study that gives computers the ability to learn without being explicitly programmed [81].…”

Section: As a New Synthetic Approachmentioning

confidence: 99%

New trends in nonconventional carbon dot synthesis

Bartolomei

Dosso

Prato

2021

Trends in Chemistry

View full text Add to dashboard Cite

Carbon dots (CDs) are currently one of the hot topics in the nanomaterial world. Until recently, their preparation has been mostly based on solvothermal or hydrothermal syntheses requiring high temperatures, long reaction times, or toxic solvents. Moreover, the resulting materials are often affected by low reproducibility and difficult purification. A potential solution to these problems could be represented by innovative fields of chemistry, such as mechanochemistry, flow chemistry, and laser synthesis in the liquid phase. Machine learning could also be applied to go beyond the trial-and-error approach commonly used to explore the CD chemical space. In this review, we explore these recent approaches and their future potential to address some of the CD limitations, widening the range of properties and applications of these highly promising nanomaterials. Synthesis of carbon dots: current challengesCarbon dots (CDs) are currently one of the most studied carbon-based materials, due to their highly promising properties. CDs are carbon-based nanoparticles presenting dimensions <10 nm, emissions ranging over the whole visible spectrum, low toxicity, and long-term photo and colloidal stability [1,2]. Generally, CDs are synthesized following either a top-down or bottomup approach, with the latter more commonly used due to its superior versatility and accessibility [3]. In fact, CD synthesis mostly relies on hydrothermal or solvothermal treatment of relatively small molecular precursors (citric acid, amino acids, carbohydrates, anilines) involving either high temperatures/pressures, long reaction times, or toxic organic solvents [4][5][6][7][8][9][10][11]. This approach is highly appreciated for the simple and economic set up and for the wide range of precursors that can be employed. An improvement in solvothermal synthesis has been achieved by using microwaves. This resulted in a great reduction of reaction times and solvent amount, mainly arising from a more controllable thermal transport [4,12]. Despite these advantages, the high temperatures needed (>150°C) result in multiple reaction pathways that are not clearly predictable, leading to extensive formation of by-products, which are often highly emissive (Figure 1) [13][14][15][16][17]. As a result, various papers identified the presence of molecular fluorophores in CD samples, highlighting the urgent need for more stringent purification protocols [18][19][20][21]23,25]. This aspect, together with irregular heat and mass transfer typical of batch synthesis, often contributes to the low reproducibility observed in CD preparation [4,22]. Furthermore, the procedures used to tailor the obtained materials are still based on a trial-and-error strategy. This approach is limited by the absence of predictability, resulting in long and difficult optimizations of the target properties and extensive time consumption. To go beyond these limitations, approaches able to increase the reproducibility of the synthesis and finely tailor the material characteristics are required. A poss...

show abstract

Molecular Machine Learning: The Future of Synthetic Chemistry?

Cited by 58 publications

References 47 publications

A review on machine learning algorithms for the ionic liquid chemical space

A review on machine learning algorithms for the ionic liquid chemical space

FragNet, a Contrastive Learning-Based Transformer Model for Clustering, Interpreting, Visualizing, and Navigating Chemical Space

New trends in nonconventional carbon dot synthesis

Contact Info

Product

Resources

About