1. RECOB: Reliable Benchmarking of Experimental Optimization in Chemistry and Materials Science1
The unreviewed RECOB study benchmarks optimisation on 14 single-objective and two multi-objective tasks built from physical chemistry experiments. Learned response models and replay of measured tables preserve broadly consistent method rankings. This provides a more experimentally grounded test of optimiser choice than synthetic functions alone. It remains a benchmark comparison, not a new prospective laboratory campaign.1
2. Collaborative Laboratory Edge Networks2
The unreviewed ChemFedFusion study combines federated chemistry-model training with task routing across heterogeneous laboratory computing sites. Benchmark tests report faster convergence and more timely, chemically valid answers. Its failure analysis nevertheless finds routes that parse correctly but are synthetically infeasible. The work connects distributed model deployment to chemical usefulness; our account is limited to the deposited abstract.2
3. Resolving Positional Markush Structures for Chemical Reaction Extraction3
An unreviewed reaction-extraction study combines learned molecular-image recognition with geometric rules for resolving positional Markush structures. On 450 reactions from 23 schemes, end-to-end F1 reached 0.46, rising to 0.54 after excluding text-dependent ambiguities. Figure reorganisation strongly affected performance. The benchmarks expose a practical bottleneck in reusable reaction data; reporting is limited to the deposited abstract.3
4. Data-Driven Exploration of Literature-Derived Catalyst and Reaction Spaces for the Vapor-Phase Aldol Condensation of Acetate with Formaldehyde4
An unreviewed study compiles 375 observations from 71 publications on catalytic aldol condensation and evaluates interpretable learning with publication-grouped validation. Sparse, heterogeneous data caused substantial overfitting, while analysis identified operational variables associated with productivity. Underreported time-on-stream data limited assessment of catalyst deactivation. The work informs literature-data reuse; our account is restricted to the deposited abstract.4
5. Predicting spatiotemporal coke growth in an industrial ethane cracker with a hybrid mechanistic–neural surrogate of its plant-tuned SPYRO model5
An unreviewed hybrid model predicts coke growth along an industrial ethane-cracker coil using mechanistic kinetics and a neural component. Six held-out operating cases showed close agreement with its plant-tuned SPYRO reference model while preserving monotonic growth. The target is simulated coke thickness, which cannot be measured in service. Reporting is limited to the deposited abstract, not independent plant validation.5
References
Xie, Zikai; Wan, Jiaming; Chen, Linjiang. RECOB: Reliable Benchmarking of Experimental Optimization in Chemistry and Materials Science. arXiv; 2026; Preprint v1; not peer reviewed. Accessed 2026-09-21T16:11:10.841Z.
Source evidence and access
Abstract and submission history, arXiv:2609.20891
Evidence paraphrase, not a quotation: The unreviewed RECOB study benchmarks optimisation on 14 single-objective and two multi-objective tasks built from physical chemistry experiments. Learned response models and replay of measured tables preserve broadly consistent method rankings. This provides a more experimentally grounded test of optimiser choice than synthetic functions alone. It remains a benchmark comparison, not a new prospective laboratory campaign.
Original arXiv abstract and submission history inspected; full text not reviewed.
Andre Williams; Shanice Brown; Kevin Thompson; Jamar Campbell; Alicia Morgan. Collaborative Laboratory Edge Networks. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15009129/v1. Accessed 2026-09-21T16:11:10.842Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15009129/v1
Evidence paraphrase, not a quotation: The unreviewed ChemFedFusion study combines federated chemistry-model training with task routing across heterogeneous laboratory computing sites. Benchmark tests report faster convergence and more timely, chemically valid answers. Its failure analysis nevertheless finds routes that parse correctly but are synthetically infeasible. The work connects distributed model deployment to chemical usefulness; our account is limited to the deposited abstract.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata from Crossref; full text not inspected.
Ulrick Fineddie Randriharimanamizara; Atif Anwer; Julien Roger; Fabrice Meriaudeau; Paul Fleurat-Lessard. Resolving Positional Markush Structures for Chemical Reaction Extraction. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15009037/v1. Accessed 2026-09-21T16:11:10.842Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15009037/v1
Evidence paraphrase, not a quotation: An unreviewed reaction-extraction study combines learned molecular-image recognition with geometric rules for resolving positional Markush structures. On 450 reactions from 23 schemes, end-to-end F1 reached 0.46, rising to 0.54 after excluding text-dependent ambiguities. Figure reorganisation strongly affected performance. The benchmarks expose a practical bottleneck in reusable reaction data; reporting is limited to the deposited abstract.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata from Crossref; full text not inspected.
Simon B. Verstraeten; Yayati Naresh Palai; Ekaterina V. Makshina; Bert F. Sels. Data-Driven Exploration of Literature-Derived Catalyst and Reaction Spaces for the Vapor-Phase Aldol Condensation of Acetate with Formaldehyde. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15008977/v1. Accessed 2026-09-21T16:11:10.842Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15008977/v1
Evidence paraphrase, not a quotation: An unreviewed study compiles 375 observations from 71 publications on catalytic aldol condensation and evaluates interpretable learning with publication-grouped validation. Sparse, heterogeneous data caused substantial overfitting, while analysis identified operational variables associated with productivity. Underreported time-on-stream data limited assessment of catalyst deactivation. The work informs literature-data reuse; our account is restricted to the deposited abstract.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata from Crossref; full text not inspected.
Hongtao Li; Hao Zhou; Jinshen Wang; Xiaochen Sun. Predicting spatiotemporal coke growth in an industrial ethane cracker with a hybrid mechanistic–neural surrogate of its plant-tuned SPYRO model. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15009054/v1. Accessed 2026-09-21T16:11:10.842Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15009054/v1
Evidence paraphrase, not a quotation: An unreviewed hybrid model predicts coke growth along an industrial ethane-cracker coil using mechanistic kinetics and a neural component. Six held-out operating cases showed close agreement with its plant-tuned SPYRO reference model while preserving monotonic growth. The target is simulated coke thickness, which cannot be measured in service. Reporting is limited to the deposited abstract, not independent plant validation.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata from Crossref; full text not inspected.
Publication record
Published 2026-09-21.
Sources, selection and claims were checked in an independent AI editorial review, followed by the AI editor's approval. This is not academic peer review.
- 2026-09-21 · Version 18a0630e · Viewing this version