MolDeBERTa-base-10M-mlm

MolDeBERTa-base-10M-mlm

This model corresponds to the MolDeBERTa base architecture pretrained on the 10M dataset using the MLM pretraining objective.

[Paper] | [Github Repo] | [Dataset on HuggingFace] | [Model Collection] | [Cite]

Abstract

Foundational models that learn the "language" of molecules are essential for accelerating material and drug discovery. These self-learning models can be trained on large collections of unlabelled molecules, enabling applications such as property prediction, molecule design, and screening for specific functions. However, existing molecular language models rely on masked language modeling, a generic token-level objective that is agnostic to physicochemical and substructure molecular properties. Here we introduce MolDeBERTa, a chemistry-informed self-supervised molecular encoder built upon the DeBERTaV2 architecture with byte-level Byte-Pair Encoding (BPE) tokenization. MolDeBERTa is pretrained on up to 123 million SMILES from PubChem using three novel pretraining objectives designed to inject strong inductive biases for molecular properties and substructure similarity directly into the latent space. The model is systematically investigated across three architectural scales, two dataset sizes, and five distinct pretraining objectives, of which three are novel and two are adapted from prior work. When evaluated on 9 MoleculeNet benchmarks, MolDeBERTa achieves the best overall performance on 4 out of 9 tasks and outperforms SMILES-based encoders on 7 out of 9 tasks, with up to a 16% reduction in regression error, and improvements of up to 2.2 ROC-AUC points on classification tasks.

Model Details

MolDeBERTa is a family of encoder-based molecular foundation models built upon the DeBERTaV2 encoder architecture and pretrained on large-scale SMILES data. The framework was evaluated across three architectural scales (tiny, small, and base), pretrained on two datasets of substantially different sizes (10M and 123M molecules), and optimized using five distinct pretraining objectives, resulting in a total of 30 pretrained model variants.

Model Selection Guide

Unsure which of the 30 models to use for your task? Based on our benchmark results, we recommend the following configurations:

• Tiny Models: MolDeBERTa-tiny-123M-contrastive_mtr or MolDeBERTa-tiny-123M-mtr
• Small Models: MolDeBERTa-small-123M-contrastive_mlc or MolDeBERTa-small-123M-contrastive_mtr
• Base Models: MolDeBERTa-base-123M-contrastive_mtr or MolDeBERTa-base-123M-mtr (Overall SOTA)

Model Usage

You can use this model for feature extraction (embeddings) or fine-tune it for downstream prediction tasks (such as property prediction or sequence classification). The embeddings may be used for similarity measurements, visualization, or training predictor models.

Usage Example

Use the code below to get started with the model:

import torch
from transformers import AutoModel, AutoTokenizer

# Load the model and tokenizer
model_name = "SaeedLab/MolDeBERTa-base-10M-mlm"
model = AutoModel.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Example input
smiles = ["CC(=O)Oc1ccccc1C(=O)O"]

# Tokenize and extract embeddings
inputs = tokenizer(smiles, return_tensors="pt", padding=True, truncation=True)
with torch.no_grad():
  outputs = model(**inputs)

# Access the last hidden state
embeddings = outputs.last_hidden_state
print(embeddings.shape)

Citation

If you use MolDeBERTa in your research, please cite:

@article{moldeberta2026,
  title   = {MolDeBERTa: Foundational Model for Physicochemical and Substructure-Informed Molecular Representation Learning},
  author  = {Oliveira, Gabriel Bianchin de and Saeed, Fahad},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.64898/2026.02.15.706011},
  url     = {https://doi.org/10.64898/2026.02.15.706011}
}

License

This model and associated code are released under the CC-BY-NC-ND 4.0 license and may only be used for non-commercial, academic research purposes with proper attribution. Any commercial use, sale, or other monetization of this model and its derivatives, which include models trained on outputs from the model or datasets created from the model, is prohibited and requires prior approval. Downloading the model requires prior registration on Hugging Face and agreeing to the terms of use. By downloading this model, you agree not to distribute, publish or reproduce a copy of the model. If another user within your organization wishes to use the model, they must register as an individual user and agree to comply with the terms of use. Users may not attempt to re-identify the deidentified data used to develop the underlying model. If you are a commercial entity, please contact the corresponding author.

Contact

For any additional questions or comments, contact Fahad Saeed ([email protected]).