1. MoTIF-X: A Multimodal Tokenized Framework for Interpretable and Extensible Molecular Representation Learning1
MoTIF-X links molecular graphs, strings and conformational information through shared chemical-motif tokens. The unreviewed model reports improved predictions across nine ADMET endpoints and tests transfer to drug–target interactions, including an external dataset of unseen drugs. Substructure attributions are compared with measured activity changes, supporting interpretable molecular representation. Our account is limited to the original abstract.1
2. A transcriptional reporting system for sensing interactions of small molecules with proteins2
An unreviewed binding sensor links small molecules to DNA and proteins to a transcription-activation domain, converting their interaction into amplified RNA output. Demonstrations cover binder characterization, molecular glues and DNA-encoded library selections. The approach could simplify measuring interactions by turning binding into a countable molecular signal. Our account is restricted to the repository-deposited abstract.2
3. De novo design of functional RNAs through higher-order interactions3
An unreviewed RNA-design framework combines higher-order interaction scoring with physics-guided sequence sampling. Designed Mango II molecules retained fluorogenic activity, while five tested twister ribozymes showed mean endpoint cleavage of 37.7–50.6%, compared with 23.5% for the wild type. These functional tests complement computational structure and sequence benchmarks. Our account is restricted to the repository-deposited abstract.3
4. AtomWeaver: Multi-Component Flow Matching with a Structured Geometric Prior Facilitates Non-Canonical Peptide Design4
AtomWeaver designs peptide side chains without fixing amino-acid identity until decoding, making noncanonical residue vocabulary expandable without retraining. The unreviewed method tests ranking against measured mutation data from two peptide–target systems and explores a 300-residue vocabulary, including types unseen during training. These results support flexible sequence design, rather than a new prospective binding campaign. Our account is abstract-only.4
5. A biological-response compound representation allows chemical perturbation prediction across cell lines5
BioPert represents compounds through measured gene-expression responses in a reference cell line, then predicts responses in other cellular contexts. The unreviewed study outperforms structure-based representations and simply copying the reference response across two datasets. Performance varies with experimental reproducibility, highlighting batch effects as a practical boundary. The approach could reduce repeated screening; our account is limited to the original abstract.5
References
Linqing Mo, Jiayu Zhou, Bin Chen. MoTIF-X: A Multimodal Tokenized Framework for Interpretable and Extensible Molecular Representation Learning. arXiv; 2026; Unreviewed preprint, version 1; arXiv-issued DOI via DataCite, pending registration. DOI: 10.48550/arXiv.2609.37384. Accessed 2026-10-01T16:33:35.634Z.
Source evidence and access
arXiv v1 abstract and Submission history
Paraphrase of inspected original abstract: MoTIF-X links molecular graphs, strings and conformational information through shared chemical-motif tokens. The unreviewed model reports improved predictions across nine ADMET endpoints and tests transfer to drug–target interactions, including an external dataset of unseen drugs. Substructure attributions are compared with measured activity changes, supporting interpretable molecular representation.
Restricted to original repository abstract and submission history; full methods and supplements not inspected.
Koder Dagher; Minqi Pan; Basilius Sauter; Saule Zhanybekova; Pinwen Cai; Lukas Schneider; Dennis Gillingham. A transcriptional reporting system for sensing interactions of small molecules with proteins. ChemRxiv; 2026; Unreviewed preprint, version 1. DOI: 10.26434/chemrxiv.15009663/v1. Accessed 2026-10-01T16:33:35.634Z.
Source evidence and access
Original repository-deposited v1 abstract and posted date in Crossref
Paraphrase of inspected original abstract: An unreviewed binding sensor links small molecules to DNA and proteins to a transcription-activation domain, converting their interaction into amplified RNA output. Demonstrations cover binder characterization, molecular glues and DNA-encoded library selections. The approach could simplify measuring interactions by turning binding into a countable molecular signal.
Restricted to repository-deposited abstract and metadata in Crossref; full paper and supplements not inspected.
Tongwei Yuan; Dong Wang; Xin-Long Chen; Han-Lin Tao; Chen-Chen Zheng; Xiao-Cong Zhao; Ya-Lan Tan; Xing-Hua Zhang; Zhi-Jie Tan. De novo design of functional RNAs through higher-order interactions. bioRxiv; 2026; Unreviewed preprint, version 1. DOI: 10.64898/2026.09.26.754601. Accessed 2026-10-01T16:33:35.634Z.
Source evidence and access
Original deposited abstract and published-online/posted date in Crossref DOI record
Paraphrase of personally inspected original abstract: An unreviewed RNA-design framework combines higher-order interaction scoring with physics-guided sequence sampling. Designed Mango II molecules retained fluorogenic activity, while five tested twister ribozymes showed mean endpoint cleavage of 37.7–50.6%, compared with 23.5% for the wild type. These functional tests complement computational structure and sequence benchmarks.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata in Crossref; full paper and supplements not inspected.
Alexander Kitaygorodsky; David Earl Hostallero; Aron Broom; Elliot Layne; Sungwon Hwang; Ashlin K. Kanawaty; Tomáš Babej; Glenn L. Butterfoss; Mark Fingerhuth. AtomWeaver: Multi-Component Flow Matching with a Structured Geometric Prior Facilitates Non-Canonical Peptide Design. bioRxiv; 2026; Unreviewed preprint, version 1. DOI: 10.64898/2026.09.23.753608. Accessed 2026-10-01T16:33:35.634Z.
Source evidence and access
Original deposited abstract and published-online/posted date in Crossref DOI record
Paraphrase of personally inspected original abstract: AtomWeaver designs peptide side chains without fixing amino-acid identity until decoding, making noncanonical residue vocabulary expandable without retraining. The unreviewed method tests ranking against measured mutation data from two peptide–target systems and explores a 300-residue vocabulary, including types unseen during training. These results support flexible sequence design, rather than a new prospective binding campaign.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata in Crossref; full paper and supplements not inspected.
Lola Le Breton; Léa Kaufmann; Elise Carraz-Billat; Quentin Fournier; Sebastien Lemieux. A biological-response compound representation allows chemical perturbation prediction across cell lines. bioRxiv; 2026; Unreviewed preprint, version 1. DOI: 10.64898/2026.09.28.755146. Accessed 2026-10-01T16:33:35.634Z.
Source evidence and access
Original deposited abstract and published-online/posted date in Crossref DOI record
Paraphrase of personally inspected original abstract: BioPert represents compounds through measured gene-expression responses in a reference cell line, then predicts responses in other cellular contexts. The unreviewed study outperforms structure-based representations and simply copying the reference response across two datasets. Performance varies with experimental reproducibility, highlighting batch effects as a practical boundary. The approach could reduce repeated screening; our account is limited to the original abstract.
Restricted to original publisher/repository-deposited abstract and bibliographic metadata in Crossref; full paper and supplements not inspected.
Publication record
Published 2026-10-01.
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.
- 2026-10-01 · Version 095d9c7a · Viewing this version