---
license: mit
library_name: pytorch
tags:
- biology
- antimicrobial-peptides
- mic-prediction
- apexoracle
---

# ApexOracle Core MIC quickstart

This repository contains one inference-only member of the ApexOracle
hierarchical MIC predictor and one text-only embedding bundle for an executable
quickstart.

The checkpoint is a project-owned model released under MIT. The example refers
to DBAASP 2136 and *Acidipropionibacterium acidipropionici* ATCC 4965; underlying
third-party records are not relicensed by MIT. See the Core repository's
`DATA_NOTICE.md` for the provenance and license boundary.

## Files

- `apexoracle_mic_strain_group0_member0_inference.pth`: strain-holdout group 0,
  ensemble member 0, stripped of optimizer and unused classification state.
- `example_text_only_dbaasp_2136_atcc_4965.pt`: precomputed molecular and
  strain-text embeddings using the public input-bundle schema.
- `example_output_cpu.json`: verified CPU output.
- `manifest.json`: file hashes, sizes, and source-checkpoint provenance.

This is a runnable single-member example, not the seven-member ensemble used
for paper-level metrics and not a prospective activity claim.

## Run

Install ApexOracle Core from its public GitHub repository, then run:

```bash
apexoracle-predict-mic \
  --checkpoint apexoracle_mic_strain_group0_member0_inference.pth \
  --input example_text_only_dbaasp_2136_atcc_4965.pt \
  --device cpu
```

The expected output is `prediction_z = -0.0717465281` and
`predicted_mic_um = 11.79631996`.