BioEmu (Microsoft)

How SubSeq Runs BioEmu

• Entrypoint is fixed to python -m bioemu.sample.
• Runtime dependencies are in the image; BioEmu source code is expected under /ref (for example /ref/src/bioemu).
• Model checkpoint/config are pinned to local paths: /ref/checkpoints/bioemu-v1.2/checkpoint.ckpt and /ref/checkpoints/bioemu-v1.2/config.yaml.
• Service-enforced paths: --output_dir=/outputs, --cache_embeds_dir=/outputs/.bioemu_embeds_cache, --cache_so3_dir=/ref/.bioemu_so3_cache.
• This deployment rejects direct sequence strings and FASTA input at submit-time; --sequence must point to an .a3m file.

Required Arguments

• --sequence: path to an .a3m file (for example /inputs/sequence.a3m).
• --num_samples: number of sampled structures to generate.

User-supplied --ckpt_path, --model_config_path, and --model_name are ignored by SubSeq.

Quick Start

--sequence=/inputs/sequence.a3m
--num_samples=1
--batch_size_100=10
--filter_samples=False
--base_seed=42

This is the same style as the BioEmu prefill in the New Job form.

Input Pattern

A3M from mounted inputs:

--sequence=/inputs/sequence.a3m
--num_samples=100
--batch_size_100=20

For this deployment profile, only .a3m input is accepted.

Outputs and Caches

• Main trajectory output is written under /outputs (for example samples.xtz plus metadata files).
• Embedding cache is written to /outputs/.bioemu_embeds_cache.
• SO(3) lookup cache is written to /ref/.bioemu_so3_cache.
• BioEmu may initialize a ColabFold environment at /ref/.bioemu_colabfold for MSA-related embedding flow.

Offline/Reproducible Use

• Pre-stage BioEmu source and v1.2 checkpoint/config under /ref.
• Prefer A3M input when you want to avoid remote MSA services.
• After warm-up caches are created, you can tighten network and mount policies in your deployment flow if desired.