Yoked learning in molecular data science DOI Creative Commons

Zhixiong Li,

Yan Xiang,

Yujing Wen

et al.

Artificial Intelligence in the Life Sciences, Journal Year: 2023, Volume and Issue: 5, P. 100089 - 100089

Published: Dec. 2, 2023

Active machine learning is an established and increasingly popular experimental design technique where the model can request additional data to improve model's predictive performance. It generally assumed that this optimal for since it relies on predictions or architecture therefore cannot be transferred other models. Inspired by research in pedagogy, we here introduce concept of yoked a second learns from selected another model. We found 48% benchmarked combinations, performed similar better than active learning. analyze distinct cases which In particular, prototype Yoked Deep Learning (YoDeL) classic provides deep neural network, thereby mitigating challenges such as slow refitting time per iteration poor performance small datasets. summary, expect new (deep) provide competitive option boost benefit capabilities multiple models during acquisition, training, deployment.

Language: Английский

Leveraging bounded datapoints to classify molecular potency improvements DOI Creative Commons
Zachary Fralish,

Paul Skaluba,

Daniel Reker

et al.

RSC Medicinal Chemistry, Journal Year: 2024, Volume and Issue: 15(7), P. 2474 - 2482

Published: Jan. 1, 2024

We present a novel data pre-processing approach, “DeltaClassifier”, that enables classification models to access traditionally inaccessible bounded datapoints guide molecular optimizations by directly contrasting pairs of molecules.

Language: Английский

Citations

4
32nd Annual GP2A Medicinal Chemistry Conference DOI Creative Commons
Jean‐Jacques Hélesbeux, Florence O. McCarthy, Maria Manuel Silva

et al.

Drugs and Drug Candidates, Journal Year: 2025, Volume and Issue: 4(1), P. 2 - 2

Published: Jan. 9, 2025

The Group for the Promotion of Pharmaceutical Chemistry in Academia (GP2A) held its 32nd annual conference August 2024 at University Coimbra, Portugal. There were 8 keynote presentations, 12 early career researcher oral and 34 poster presentations. Four awards delivered, two best communications

Language: Английский

Citations

0
Unveiling New Reactivities in Complex Mixtures: Synthesis of Tricyclic Pyridinium Derivatives DOI Creative Commons
Johanan Kootstra, Jaya Mehara, Marieke J. Veenstra

et al.

Journal of the American Chemical Society, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 29, 2025

The discovery of new transformations drives the development synthetic organic chemistry. While main goal chemists is to obtain maximum yield a desired product with minimal side formation, meticulous characterization latter offers an opportunity for discovering reaction pathways, alternative mechanisms, and products. Herein, we present case study on chemical transformation using online mass spectrometry. This highly sensitive method enabled pathway in catalyst-free cross-dehydrogenative coupling 1,2,3,4-tetrahydroisoquinoline acetone via peroxide intermediate, ultimately yielding tricyclic pyridinium compound. Mass spectrometry was instrumental detecting identifying structure compound, initially formed as trace byproduct, which allowed us develop general methodology its exclusive formation.

Language: Английский

Citations

0
Reaction development: a student's checklist DOI
Jasper L. Tyler, Dirk Trauner, Frank Glorius

et al.

Chemical Society Reviews, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

So you've discovered a reaction. This review discusses the key areas involved in developing new reactions and provides handy checklist guide to help maximise potential of your novel transformation.

Language: Английский

Citations

0
Data-Driven Search Algorithm for Discovery of Synthesizable Zeolitic Imidazolate Frameworks DOI Creative Commons
Soochan Lee,

Hyein Jeong,

Sungyeop Jung

et al.

JACS Au, Journal Year: 2025, Volume and Issue: 5(3), P. 1460 - 1470

Published: March 7, 2025

Zeolitic imidazolate frameworks (ZIFs), metal–organic analogues of zeolites, hold great potential for carbon-neutral applications due to their exceptional stability and porosity. However, ZIF discovery has been hindered by the limited topologies resulting from a mismatch between numerous predicted few synthesized zeolitic networks. To address this, we propose data-driven search algorithm using structural descriptors known materials as screening tool. From over 4 million zeolite structures, identified candidates based on O–T–O angle differences, vertex symbols, T–O–T angles. Energy calculations facilitated ranking ZIFs synthesizability, leading successful synthesis three with two novel topologies: UZIF-31 (uft1) UZIF-32, -33 (uft2). Notably, UZIF-33 exhibited remarkable CO2 selective adsorption. This study highlights synergistic combining predictions chemical intuition advance material discovery.

Language: Английский

Citations

0
Chiral nanographenes exhibiting circularly polarized luminescence DOI Creative Commons
V. Ravi Kumar,

J. Páez,

Sandra Míguez‐Lago

et al.

Chemical Society Reviews, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Chiral nanographenes: molecular architecture, key constituents, and their circularly polarized luminescence.

Language: Английский

Citations

0
Uncovering Zeolitic Imidazolate Frameworks through Digitization of Chemical Insights DOI Creative Commons
Soochan Lee,

Hyein Jeong,

Sungyeop Jung

et al.

Published: April 12, 2024

The vast number of computational predictions presents challenges when transitioning from structural models to experimental confirmations. To address this challenge, we digitized chemical intuition into the discovery process, focusing on zeolitic imidazolate frameworks (ZIFs). Despite their potential, limited topologies by “zeolite conundrum” and an unclear synthetic roadmap have hindered ZIF discovery. We propose a data-driven approach for using descriptors known materials as screening tool. From over 4 million zeolite structures, identified potential candidates based O−T−O angle differences, vertex symbols, T−O−T angles. Energy calculations enabled ranking synthesizability ZIFs, resulting in successful synthesis three ZIFs with two unprecedented topologies, UZIF-31 (uft1) UZIF-32, -33 (uft2). Notably, UZIF-33 demonstrated remarkable selective adsorption CO2. This work underscores synergistic combining structure advance field material

Language: Английский

Citations

1
Data-Driven Search Algorithm for Discovery of Synthesizable Zeolitic Imidazolate Frameworks DOI Creative Commons
Soochan Lee,

Hyein Jeong,

Sungyeop Jung

et al.

Published: Aug. 6, 2024

Zeolitic imidazolate frameworks (ZIFs), metal-organic analogues of zeolites, hold great potential for carbon-neutral applications due to their exceptional stability and porosity. However, ZIF discovery has been hindered by the limited topologies resulting from a mismatch between numerous predicted few synthesized zeolitic networks. To address this, we propose data-driven search algorithm using structural descriptors known materials as screening tool. From over 4 million zeolite structures, identified candidates based on O−T−O angle differences, vertex symbols, T−O−T angles. Energy calculations facilitated ranking ZIFs synthesizability, leading successful synthesis three with two novel topologies: UZIF-31 (uft1) UZIF-32, -33 (uft2). Notably, UZIF-33 exhibited remarkable CO2 selective adsorption. This study highlights synergistic combining predictions chemical intuition advance material discovery.

Language: Английский

Citations

1
Finding the most potent compounds using active learning on molecular pairs DOI Creative Commons
Zachary Fralish, Daniel Reker

Beilstein Journal of Organic Chemistry, Journal Year: 2024, Volume and Issue: 20, P. 2152 - 2162

Published: Aug. 27, 2024

Active learning allows algorithms to steer iterative experimentation accelerate and de-risk molecular optimizations, but actively trained models might still exhibit poor performance during early project stages where the training data is limited model exploitation lead analog identification with scaffold diversity. Here, we present ActiveDelta, an adaptive approach that leverages paired representations predict improvements from current best compound prioritize further acquisition. We apply ActiveDelta concept both graph-based deep (Chemprop) tree-based (XGBoost) exploitative active for 99 K

Language: Английский

Citations

1
Digital Chemistry: Navigating the Confluence of Computation and Experimentation – Definition, Status Quo, and Future Perspective DOI Creative Commons

Stefan Bräse

Digital Discovery, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 1, 2024

Digital chemistry represents a transformative approach integrating computational methods, digital data, and automation for chemical sciences. toolkits were used to simulate, predict, accelerate, analyze processes properties.

Language: Английский

Citations

0