deepmirror

deepmirror is a command-line interface for interacting with the deepmirror API. It allows you to train models, run predictions, and submit structure prediction jobs directly from your terminal.

Installation

pip install deepmirror

Authentication

Before using most commands, you need to log in to get your API token:

dm login EMAIL

This saves your token and host in ~/.config/deepmirror/ for reuse.

Model Commands

List Available Models

dm model list

View Model Metadata

dm model metadata MODEL_ID

Get Full Model Info

dm model info MODEL_ID

Train a Custom Model

dm train --model-name mymodel \
  --csv-file path/to/data.csv \
  --smiles-column smiles \
  --value-column target \
  [--classification]

• --classification enables classification mode.
• Default SMILES column is smiles, target column is target.

Run Inference

You can run inference using either a CSV file or direct SMILES input:

# From a CSV or TXT file
dm predict --model-name mymodel --csv-file inputs.csv

# Direct SMILES
dm predict --model-name mymodel --smiles "CCO"

Batch Inference

Upload a Parquet file for large-scale predictions:

dm batch create MODEL_ID path/to/input.parquet

Check job status and download results once completed:

dm batch status TASK_ID
dm batch download TASK_ID predictions.parquet

Co-folding and Affinity Predictions

Explore co-folding capabilities using the following notebooks:

• Predict 3D structures of protein–ligand complexes via co-folding

• Apply constraints during co-folding to guide the predicted structure

• Estimate binding affinity values alongside structural prediction using Boltz-2

• Apply Covalent Bond Constraints to guide co-folding of covalent ligands in Chai-1

💡 Tips

• If a token is missing or expired, commands will prompt you to log in again.

• Use --help on any command for more details, e.g.:

  dm train --help