# LA4SR TI-inclusive algaGPT Distribution

This distribution contains the TI-inclusive algaGPT model packaged for easy use with Singularity, along with datasets, inference scripts, and evaluation tools.

## Directory Structure

* **`la4sr_sp2.sif`** (6.8 GB): Singularity container with the complete computational environment.
* **Model Files:**

  * `ckpt.pt`: Model checkpoint file (990 MB).
  * `meta.pkl`: Metadata for the model.
  * `model.py`: Python script defining the model architecture.
* **Datasets:**

  * FASTA files (`generated_prompts_*_headed.fa`) for various taxa (algae, archaea, bacteria, fungi, viruses).
* **Scripts:**

  * `run_la4sr_TI-inc-algaGPT.sh`: Main inference and metrics generation script.
  * `run_la4sr_loop.sbatch`: SLURM batch script to run multiple inference jobs.
  * `infer_TI-inc-algaGPT.py`: Python inference script.
  * `llm-metrics-two-files.py`: Generates classification metrics and visualizations.
* **Utility Files:**

  * `filelist.txt`, `contam-filelist.txt`, `algae-filelist.txt`: Lists of FASTA files to analyze.
  * `la4sr_sp2.sif.md5`: Checksum for container verification.

### Subdirectories:

* **`cache/`**: Cache directory for Hugging Face models and tokenizers.
* **`results-archive/`**: Archived results from previous runs.
* **`algaGPT_fungi-algae2x-update_cleaned/`**: Fine-tuned algaGPT variant optimized for fungi.
* **`TI-free-la4sr/`**: Pythia-based TI-free flagship model.
* **`slurm-logs/`**: SLURM job output logs.

## Quick Start

1. **Setup:** Ensure Singularity is installed on your HPC or local system.

2. **Inference (no scheduler):**

   ```bash
   ./run_la4sr_TI-inc-algaGPT.sh resume <algal_fasta> <contaminant_fasta>
   ```

   Replace `<algal_fasta>` and `<contaminant_fasta>` with your FASTA file paths.

3. **Inference (with SLURM scheduler):**

   * Update `algae-filelist.txt` and `contam-filelist.txt` with paths to your FASTA files.
   * Submit the SLURM job array:

     ```bash
     sbatch run_la4sr_loop.sbatch
     ```

## Outputs

Results, including TSV files, metrics reports, misclassification reports, and visualizations, are stored in the `results/` directory.

## Additional Information

* To manually run the inference script:

  ```bash
  singularity exec --nv la4sr_sp2.sif python3 infer_TI-inc-algaGPT.py --init_from resume input.fasta -o output.tsv
  ```

* To generate metrics independently:

  ```bash
  singularity exec la4sr_sp2.sif python3 llm-metrics-two-files.py algal_results.tsv contaminant_results.tsv -o metrics_report.txt -m misclassified_report.txt -v
  ```

---

For further assistance, contact the maintainers.

The algaGPT model was trained by Kourosh Salehi-Ashtiani, Ph.D. with help in training data preparation from Ashish Kumar Jaiswal, Msc.