Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 49 min ago
PreprintCC BYarXiv (Cornell University)

Mol-JEPA: A multimodal Joint Embedding Predictive Architecture for Molecules

  • Florian Rottach1,2
  • Sebastian Schieferdecker1
  • William Rudman3
  • Randall Balestriero4
  • Carsten Eickhoff2
4 affiliations
  1. 1Boehringer Ingelheim
  2. 2University of Tübingen
  3. 3The University of Texas at Austin
  4. 4Brown University
Open at sourcePDFDOICite

Abstract

Despite recent advances in molecular foundation models, several limitations remain, such as chemically invalid augmentations, modality collapse, and incomplete representation of biochemical environments. To address these challenges, we present \textbf{Mol-JEPA}, a scalable framework for learning molecular world models. Rather than relying on suboptimal molecular perturbations, our model uses modality masking to exploit information from molecular structures, cellular phenotypes, binding affinities, ADMET profiles, quantum chemistry simulations and other drug discovery data. Across various benchmarks, we show that the representations learned by Mol-JEPA deliver strong performance, demonstrating the value of incorporating biochemical context through latent space prediction.

Related

  1. Multimodal computer vision and machine learning for predictive monitoring of planar-flow casting of amorphous steel foils
    Ritaban Dutta, Michael Kellam, Daniel Liang
    PLOS OnePeer-reviewedCC BY 4.0Full textNeuroscienceMaterials & polymersPLOSPMC
  2. From Static Material Design to Predictive Open-Loop Control via Chemistry–Architecture Coupling in a Moving-Boundary Reaction–Diffusion Framework
    Andualem Belachew Workie
    ChemRxivPreprintCC BYFull textGeneral chemistryChemRxiv
  3. Scalable Density-Based Bootstrap Embedding with Perturbative Triples for Large Molecular Systems
    Yi Sun
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
  4. Generalized local potential functional embedding theory of localized orbitals
    Wafa Makhlouf, Bruno Senjean, Emmanuel Fromager
    Electronic StructurePeer-reviewedLicence not statedFull text1 citedTheoretical & computational chemistryHAL
  5. POLEN: An OpenMM Framework for ML/MM Simulations with Mutually Polarizable AMOEBA Embedding
    Renato D. Cunha, Leonardo J. Duarte, Maurício D. Coutinho-Neto +2
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv

Cite

BibTeXRIS
BibTeX
@misc{rottach2026moljepa,
  title = {{Mol-JEPA: A multimodal Joint Embedding Predictive Architecture for Molecules}},
  author = {Florian Rottach and Sebastian Schieferdecker and William Rudman and Randall Balestriero and Carsten Eickhoff},
  howpublished = {arXiv (Cornell University)},
  year = {2026},
  month = aug,
  doi = {10.48550/arxiv.2608.22642},
  url = {https://arxiv.org/abs/2608.22642},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - Mol-JEPA: A multimodal Joint Embedding Predictive Architecture for Molecules
AU  - Florian Rottach
AU  - Sebastian Schieferdecker
AU  - William Rudman
AU  - Randall Balestriero
AU  - Carsten Eickhoff
PY  - 2026
DA  - 2026/08/23
PB  - arXiv (Cornell University)
DO  - 10.48550/arxiv.2608.22642
UR  - https://arxiv.org/abs/2608.22642
AB  - Despite recent advances in molecular foundation models, several limitations remain, such as chemically invalid augmentations, modality collapse, and incomplete representation of biochemical environments. To address these challenges, we present \textbf{Mol-JEPA}, a scalable framework for learning molecular world models. Rather than relying on suboptimal molecular perturbations, our model uses modality masking to exploit information from molecular structures, cellular phenotypes, binding affinities, ADMET profiles, quantum chemistry simulations and other drug discovery data. Across various benchmarks, we show that the representations learned by Mol-JEPA deliver strong performance, demonstrating the value of incorporating biochemical context through latent space prediction.
SN  - 2331-8422
ER  -