Full papers
Submissions must use the NeurIPS 2026 workshop LaTeX template, with a maximum of 5 content pages (including figures and tables); references and appendices do not count toward the limit.
Machine Learning/CORE Unranked
ML4Molecules 2026
NeurIPS 2026 Workshop on Agentic Systems for Molecular Sciences
ML4Molecules 2026 is a NeurIPS workshop in Paris focusing on agentic systems in molecular sciences, exploring both the promise and limitations of AI-driven approaches in chemistry, biology, and materials science. It emphasizes rigorous evaluation, negative results, and benchmark contributions alongside methodological advances, bridging machine learning with experimental science.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Poster presentations
All accepted contributions are presented as posters.
Contributed talks
A subset of posters are selected for contributed talks, with slots reserved for early-career first authors.
Best Paper Award submission
The strongest overall contribution is eligible for a Best Paper Award.
Research areas in scope
01
Agentic Systems and Tool Use
Agentic and multi-agent systems for the molecular sciencestool use, planning, and orchestrationClosed-loop / self-driving labs, active learning, and Bayesian optimization for experimental designGrounding LLMs via chemical databases, simulators, and experimental feedback
02
Benchmarking and Evaluation
Benchmarks, reproducibility, negative results, and rigorous evaluation of scientific agents and their componentsContributions that clarify what current methods can and cannot do, including negative results, careful ablations, and rigorous baselines
03
Foundational Models and Representation Learning
Foundation models for chemistry, biology, and materials, and their benchmarkingRepresentation learning and generative models for molecules, proteins, reactions, and materialsMachine-learned interatomic potentials, force fields, molecular dynamics, and geometric / equivariant / physics-informed deep learning
04
Predictive Modeling and Applications
Bioactivity, ADME, toxicity, and molecular property predictionReaction prediction, retrosynthesis, and synthesis planningStructure-based and ligand-based virtual screening; docking, scoring, and protein–ligand / protein–protein interaction predictionPerturbation modelling, batch effect correction, domain adaptation and single-cell prediction
Policies worth checking twice
- Submissions must be double-blind anonymized: author names, affiliations, and acknowledgments must be removed, and prior work cited in the third person.
- Main text is limited to 5 content pages; references and appendices do not count toward the limit, but reviewers are not obliged to read supplementary material.
- Maximum file size is 50 MB.
- The workshop is non-archival: papers under review elsewhere are welcome, and accepted papers may be published elsewhere afterward.
- Work already published at NeurIPS or other archival ML venues should not be submitted; substantial extensions of prior non-ML-venue work are eligible.
- The NeurIPS checklist is not required.
- Submissions are managed via OpenReview, and authors must maintain up-to-date OpenReview profiles for conflict-of-interest management.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026