About - Registration - Schedule - Keynote Speakers - Keynote Abstracts - Accepted Contributions - Important Dates - Call for Papers
IMPORTANT
2024-09-18: Registration for the workshop is now open. Please register and join us!
About
Machine Learning models for molecules and materials are essential drivers in drug discovery, materials design, environmental science, and precision medicine. With these models, experts are enabled to meaningfully interact with molecular systems and to design them. The interaction with molecular systems allows for deeper insights into typically complex systems and facilitate (enable) the development of new drugs, personalized medical treatments, and innovative materials - e.g., advanced fuel cells in the field of energy conversion and storage or materials for environmental remediation in the field of environmental science. Recently, large Language Models (LLMs) have demonstrated remarkable performance across various domains in machine learning, revolutionizing natural language processing tasks and showing notable capabilities in understanding and generating human-like text. LLMs have also had a profound impact on molecular machine learning because they allow for human-AI interaction, interpretability, zero- and few-shot adaption, and in-context learning. However, the application of LLM architectures to molecular data presents significant challenges. Therefore, this workshop focuses on exploring potentials of ML methods with broader capabilities like LLMs incorporate them. We welcome all work which might provide potential steps towards these broader capabilities, e.g., by tackling the areas of knowledge and interaction, adaptability and robustness, abstraction and reasoning, or efficiency.
Join us at our workshop at which experts from diverse fields, ranging from ML and LLMs to molecular sciences, will collaborate to explore new potentials of ML methods for molecules and materials in the era of LLMs.
Registration
The workshop will be open to everyone without a registration fee. You can register here!
Schedule
Fri, Dec 6th, 09:00 am - 6:00 pm, CET; venue: Fritz Haber Institute of the Max Planck Society, Berlin; online at Zoom
| CET | Event | Speakers | Title |
|---|---|---|---|
| 08:00 - 09:00 | Registration | ||
| 09:00 - 09:30 | Invited Talk | Rocio Mercado | Generative AI for Molecular Design: From Drugs to Sustainable Materials |
| 09:30 - 10:00 | Invited Talk | Daniel Probst | Of graphs, sets, and molecules |
| 10:00 - 10:15 | Contributed Talk | Nawaf Alampara | Task Alignment Outweighs Framework Choice in Scientific LLM Agents |
| 10:15 - 10:30 | Contributed Talk | Riccardo Tedoldi | WEISS: Wasserstein efficient sampling strategy for LLMs in drug design |
| 10:30 - 11:00 | Coffee Break | ||
| 11:00 - 11:30 | Invited Talk | Marwin Segler | Deep Learning for Molecules: The First Decade |
| 11:30 - 11:45 | Contributed Talk | Nikhil Branson | AntiDIF: Accurate and Diverse Antibody Specific Inverse Folding with Discrete Diffusion |
| 11:45 - 12:00 | Contributed Talk | Yujia Guo | Bridging Data-Driven and Expert Knowledge for Interpretable Evaluation of Synthetic Routes |
| 12:00 - 12:30 | Panel Discussion | Günter Klambauer, José Miguel Hernández Lobato, TBD | |
| 12:30 - 13:30 | Lunch | ||
| 13:30 - 14:00 | Invited Talk | Nadine Schneider | Applying AI/ML to Accelerate the DMTA Cycle in Drug Discovery |
| 14:00 - 14:15 | Contributed Talk | Julian Cremer | FLOWR.root: A flow matching based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction |
| 14:15 - 14:30 | Contributed Talk | Rasmus Hannibal Tirsgaard | Semi-Supervised Learning for Molecular Graphs via Ensemble Consensus |
| 14:30 - 14:45 | Contributed Talk | Xuan Vu Nguyen | Synthelite: Chemist-aligned and feasibility-aware synthesis planning with LLMs |
| 14:45 - 15:00 | Contributed Talk | Marcel Hiltscher | Explaining What Matters: Faithfulness in Molecular Deep Learning |
| 15:00 - 15:30 | Coffee Break | ||
| 15:30 - 16:00 | ELLIS UnConference Welcome | ||
| 16:00 - 16:30 | Retreat of Program Fellows | ||
| 16:00 - 18:00 | Poster session | ||
| 18:00 - 20:00 | Reception |
Keynote Speakers
Accepted contributions (poster)
| ID | Title | Authors |
|---|---|---|
| 1 | DEQuify your force field: More efficient simulations using deep equilibrium models | Andreas Burger, Luca Thiede, Alan Aspuru-Guzik, Nandita Vijaykumar |
| 2 | ML Force Fields for Computational NMR Spectra of Dynamic Materials across Time-Scales | Lars Leon Schaaf, Benjamin J. Rhodes, Mary E. Zick, Suzi M. Pugh, Jordon S. Hilliard, Shivani Sharma, Casey R. Wade, Phillip J. Milner, Gabor Csanyi, Alexander C. Forse |
| 3 | DrugDiff - small molecule diffusion model with flexible guidance towards molecular properties | Marie Oestreich, Matthias Becker |
| 4 | Generative Model for Synthesizing Ionizable Lipids: A Monte Carlo Tree Search Approach | Jingyi Zhao, Yuxuan Ou, Austin Tripp, Morteza Rasoulianboroujeni, José Miguel Hernández-Lobato |
| 5 | In silico enzyme prediction and generation by fine tuning protein language models | Marco Nicolini, Emanuele Saitto, Ruben Jimenez, Emanuele Cavalleri, Aldo Galeano, Dario Malchiodi, Alberto Paccanaro, Peter N Robinson, Elena Casiraghi, Giorgio Valentini |
| 6 | A language model assistant for biocatalysis | Yves Gaetan Nana Teukam, Francesca Grisoni, Matteo Manica |
| 7 | Tango*: Constrained synthesis planning using chemically informed value functions | Daniel P Armstrong, Zlatko Jončev, Jeff Guo, Philippe Schwaller |
| 8 | A Generative Model for the Design of Synthesizable Ionizable Lipids | Yuxuan Ou, Jingyi Zhao, Austin Tripp, Morteza Rasoulianboroujeni, José Miguel Hernández-Lobato |
| 9 | The Jungle of Generative Drug Discovery: Traps, Treasures, and Ways Out | Rıza Özçelik, Francesca Grisoni |
| 10 | BNEM: A Boltzmann Sampler Based on Bootstrapped Noised Energy Matching | RuiKang OuYang, Bo Qiang, José Miguel Hernández-Lobato |
| 11 | Efficient and Unbiased Sampling of Boltzmann Distributions via Consistency Models | Fengzhe Zhang, Jiajun He, Laurence Illing Midgley, Javier Antoran, José Miguel Hernández-Lobato |
| 12 | UPT++: Latent Point Set Neural Operators for Modeling System State Transitions | Andreas Fürst, Florian Sestak, Artur Petrov Toshev, Benedikt Alkin, Nikolaus A. Adams, Andreas Mayr, Günter Klambauer, Johannes Brandstetter |
| 13 | Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences | Niklas Schmidinger, Lisa Schneckenreiter, Philipp Seidl, Johannes Schimunek, Pieter-Jan Hoedt, Johannes Brandstetter, Andreas Mayr, Sohvi Luukkonen, Sepp Hochreiter, Günter Klambauer |
| 14 | Enhancing Molecular Property Prediction with GNNs via Architecture-Agnostic Graph Transformations | Zhifei Li, Gerrit Großmann, Verena Wolf |
| 15 | Fantastic SMILES Augmentation Strategies and Where to Find Them | Helena Brinkmann, Francesca Grisoni, Antoine Argante, Hugo ter Steege |
| 16 | Rectified Flow For Structure Based Drug Design | Daiheng Zhang, Chengyue Gong, Qiang Liu |
| 17 | Equivariant conditional diffusion model for exploring the chemical space around Vaska’s complex | François R J Cornet, Pratham Deshmukh, Bardi Benediktsson, Mikkel N. Schmidt, Arghya Bhowmik |
| 18 | Sample Efficient Goal-Directed Catalyst Design for the Morita-Baylis-Hillman Reaction | Sarina Kopf, Jeff Guo, Jaime Martin, Johannes Schoergenhumer, Cristina Nevado, Philippe Schwaller |
| 19 | Steering generative deep learning with task negation for de novo drug design | Sarah de Ruiter, Rıza Özçelik, Francesca Grisoni |
| 20 | MØDRCN – Open-Source Chemical Reservoir Computing Tool | Mehmet Aziz Yirik, Jakob Lykke Andersen, Rolf Fagerberg, Daniel Merkle |
| 21 | Enhancing Generalizability in Permeability Prediction for Macrocyclic Peptides | Anna Borisova, Rebecca Manuela Neeser, Philippe Schwaller |
| 22 | An exploration of dataset importance in single-step retrosynthesis prediction | Sara Tanovic, Fernanda Duarte |
| 23 | Towards transparent ML for kinase drug discovery: Insights from structural binding models | Joschka Groß, Michael Backenköhler, Paula Linh Kramer, Verena Wolf, Andrea Volkamer |
| 24 | Autonomous Atomistic Simulations with Hierarchical LLM Agents | Ziqi Wang, Hongshuo Huang, Hancheng Zhao, Changwen Xu, Jan Janssen, Venkatasubramanian Viswanathan |
| 25 | Protein Language Model-Driven Zoonotic Risk Assessment of SARS-CoV-2 Variants | Payel Das, Alessandra Toniato, Aurelie Lozano, Vijil Chenthamarakshan |
| 26 | RxnRule: Filtering single step reaction predictions based on reaction class incompatibilities | Zlatko Jončev, Daniel P Armstrong, Victor Sabanza Gil, Ziad El Malki, Philippe Schwaller |
| 27 | Physics-Aware Diffusion Models for Micro-structure Material Design | Jacob K Christopher, Stephen Baek, Ferdinando Fioretto |
Important dates
- November 1, 2024: Deadline for submission
- Mid November, 2024: Author notification
- December 6, 2024: Workshop
Call for papers
We are calling for papers exploring machine learning for molecules and materials in the era of LLMs. Topics include (but not limited to):
- Molecular Dynamics and Machine Learning
- Generative Models for Molecule and Material Design
- Molecular Property Prediction
- Inverse Design and Optimization in Materials Science
- Molecular Simulation Enhancement with ML
- Representation Learning for Molecules and Materials
- Machine learning methods for chemical reactions and synthesis
- Uncertainty Quantification and Robustness in Molecular ML Models
- Active Learning and Data-Efficient ML for Molecular Science
- Language models in chemistry
Please submit your contributions on OpenReview until November 1 2024 11:59 PM UTC-0. The submissions should be in PDF and follow the NeurIPS template with a maximum of 5 pages (not including references and appendices). We also welcome extended abstract submissions (2 pages). Please anonymize your paper since the review process is dual-anonymous. Filling the NeurIPS checklist with the paper is not compulsory.
Note that workshop papers often represent the current status of on-going projects. They should not be considered as final versions of record for a particular project. The workshop is non-archival.
Best Paper Award
The best paper award (1000€) is sponsored by AstraZeneca.
Two papers share the best paper award:
- DEQuify your force field: More efficient simulations using deep equilibrium models. Andreas Burger et al.
- Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences. Niklas Schmidinger et al.
Award committee: Andrea Volkamer, Philippe Schwaller, Cecilia Clementi
Organizing Committee and Contact
Chairs: Johannes Margraf, Francesca Grisoni, Günter Klambauer
Organizing committee: Alaa Bessadok, Sovhi Luukkonen, Johannes Schimunek, Karsten Reuter, Giulia Glorani, Francesca Grisoni, Johannes Margraf, Günter Klambauer
AstraZeneca has provided a sponsorship grant towards this independent Programme.
Contact: ml4molecules@ml.jku.at