bolero
Predicting cell-state-specific gene regulation from DNA sequence.
Bolero is a cell-state-conditioned sequence-to-function model. It takes a 524,288 bp one-hot DNA sequence plus an atlas-scale cell-state embedding (and optional conditioning such as tissue, developmental age, or a TF-activity score) and predicts cell-state-specific chromatin accessibility and transcript abundance at 32 bp resolution (16,384 output bins).
Under the hood it is a frozen Borzoi/Flashzoi backbone with per-layer conditional LoRA adapters whose low-rank weights are generated on the fly from the cell-state embedding, so every cell state gets its own effective network. For a chosen cell state those adapters can be "collapsed" into a plain DNA→track model for fast inference and base-level attribution. Bolero is trained on Bolero-10M: 10.8M cells across 36 datasets and 6 mammals.
Getting started
New here? Start with Installation — one pixi install brings up the full
GPU stack. Then work through the tutorials below in order; each is a runnable notebook with
committed outputs.
Tutorials
A guided series that reproduces the paper's workflow end to end, from raw single cells to a trained model and its predictions.
Embedding & metacells — build the cell-state representation that conditions the model.
Meta cell AnnData & Parquet — turn fragments into the coverage database and pseudobulks.
Model training — fit Bolero on one dataset or the full multi-dataset atlas.
Prediction & variant effect — accessibility, caQTL/eQTL scoring, and DNA attribution.
Score model (Bolero-Score) — condition on a chromVAR TF-activity score (e.g. AP-1).
Cross-species — run the trained atlas on any species' genome, DNA-only, no retraining.
Beyond the model
bolero pairs with bolerodata, a lightweight
registry that maps short keys to the datasets, trained model zoo, and QTL collections behind
the paper. It is installed automatically alongside bolero.
Citation
If you use Bolero, please cite "Bolero: predicting cell-state-specific gene regulation from DNA sequence" (Hanqing Liu et al.). Full citation details will be added on publication.