LLPS-predict

LLPS propensity prediction for IDRs using ESM2 embeddings and a logistic-regression head.

Install

Choose one of the install paths below.

Recommended (NVIDIA GPU, fastest)

conda create -n LLPS-predict python=3.9
conda activate LLPS-predict
pip install torch --index-url https://download.pytorch.org/whl/cu124
pip install fair-esm
pip install -e .
python -c "import torch; print(torch.__version__); print('cuda', torch.cuda.is_available()); print('cuda_version', torch.version.cuda)"

Expected check output includes cuda True.

If you have a NVIDIA GPU and this fails, your driver/GPU/platform likely needs a different CUDA-enabled PyTorch build. Use the official selector to pick a compatible install command: https://pytorch.org/get-started/locally/

CPU fallback (Mac / no NVIDIA GPU, slower but supported)

conda create -n LLPS-predict python=3.9
conda activate LLPS-predict
pip install torch
pip install fair-esm
pip install -e .
python -c "import torch; print(torch.__version__); print('cuda', torch.cuda.is_available()); print('cuda_version', torch.version.cuda)"

Expected check output includes cuda False.

CLI Commands

After installation, two console commands are available:

• llps-predict
• llps-predict-per-res

Usage

Single sequence score:

llps-predict --sequence YGQSSYSSYGQSQNTGY

FASTA with many sequences:

llps-predict --sequence example_multi_sequences.fasta --output example_sequences_LLPS_propensities.csv

You may alter token batching if you run into memory issues:

llps-predict \
  --sequence example_multi_sequences.fasta \
  --toks_per_batch 4096 \
  --truncation_seq_length 1022 \
  --output LLPS_propensity.csv

Per-residue LLPS profile for a single sequence/FASTA entry:

llps-predict-per-res \
  --sequence example_single_sequence.fasta \
  --probe_lengths 15 25 40 \
  --stride 1 \
  --output tau_perRes_LLPS_scores.csv

Notes

• --toks_per_batch: higher is faster but uses more memory.
• --truncation_seq_length: sequences longer than this are truncated for ESM2 inference.

Export/Update LR Checkpoint

Inference uses a pure torch .pt LR checkpoint. If you retrain the sklearn LR model, export a new checkpoint with:

conda install scikit-learn=1.5.1 joblib
python scripts/export_lr_joblib_to_pt.py \
  --joblib model_development/LLPS_model_latest.joblib \
  --out model_development/LLPS_model_latest.pt

Additional Manuscript Data

• Manuscript datasets and minimal reproducibility scripts are in the companion repository:
  • https://github.com/samlobe/amyloid-predict/tree/main/manuscript_data
• DOI-backed Zenodo archive for the manuscript snapshot will be listed there:
  • https://doi.org/10.5281/zenodo.18882439

Acknowledgments

• ESM developers
• Tesei, Lindorff-Larsen et al. (paper)
• CD-CODE contributors
• Scott Shell and Joan-Emma Shea

Citation

If you use LLPS-predict, the model weights, or predictions generated by this software, please cite:

Lobo, S., Griem, L., Shell, M. S., & Shea, J.-E. amyloid-predict and LLPS-predict: Predicting phase separation propensities in the intrinsically disordered proteome. Proceedings of the National Academy of Sciences 123, e2531932123 (2026). https://doi.org/10.1073/pnas.2531932123

@article{Lobo_2026,
  title={amyloid-predict and LLPS-predict: Predicting phase separation propensities in the intrinsically disordered proteome},
  volume={123},
  number={22},
  journal={Proceedings of the National Academy of Sciences},
  publisher={National Academy of Sciences},
  author={Lobo, Samuel and Griem, Leif and Shell, M. Scott and Shea, Joan-Emma},
  year={2026},
  month={May},
  doi={10.1073/pnas.2531932123},
  url={https://doi.org/10.1073/pnas.2531932123}
}