algaGPT

Causal language model for binary classification of microalgal vs. contaminant protein sequences, built on nanoGPT. algaGPT distinguishes algal proteins from bacterial, archaeal, and fungal contaminants in metagenomic assemblies without requiring sequence homology.

Developed by Kourosh Salehi-Ashtiani, New York University Abu Dhabi.

Model Description

Property	Value
Architecture	nanoGPT (GPT-2, 12 layers, 12 heads, 768 embedding)
Task	Binary classification via next-token prediction
Mode	TI-inclusive (full-length amino acid sequences)
Training data	~58.6M protein sequences (1:1 algal:contaminant)
Algal sources	166 microalgal genomes across 10 phyla
Contaminant sources	Bacterial, archaeal, and fungal sequences from NCBI nr

Performance

Metric	Score
Output coverage	>99% of sequences across 166 microalgal genomes received predictions
Algal retention	83.3% across 222 reference genomes in algaGPT decoding mode
Bacterial-class precision	>95% in the matched benchmark
Speed vs. BLASTp	~10,701x faster

Usage

Input: Amino acid string ending with >

Output: Classification tag (algal/contaminant) via next-token prediction

# See Nelson et al. (2025) for full inference pipeline
# Batch inference: https://github.com/SarahD4/dark-whiteGPLM/blob/main/batch_inference.py

Application in ELF-NET

algaGPT served as the primary proteome extraction tool in the ELF-NET study. It classified 447.7 million proteins from a 2,357-sample mixed-source input collection; the 2,044 samples with non-empty Pfam-A results contained 231.7 million proteins classified as algal for downstream domain-environment analysis.

Related Resources

Resource	Link
ELF-NET analysis pipeline (371 scripts, 15 modules)	github.com/olympus-terminal/ELF-NET
Bidirectional XGBoost models	TARA-XGBoost-Bidirectional
VICReg joint embedding model	TARA-WorldModel-VICReg
Dark-whiteGPLM checkpoints	SarahDaakour/dark-whiteGPLM
Dark-whiteGPLM training data	SarahDaakour/dark-whiteGPLM-data
Dark-whiteGPLM code	github.com/SarahD4/dark-whiteGPLM
algaGPT-classified sequences (Data S2)	Zenodo 10.5281/zenodo.18728836

Authors

David R. Nelson, Ashish Kumar Jaiswal, Noha Samir Ismail, Alexandra Mystikou, Kourosh Salehi-Ashtiani

New York University Abu Dhabi

Citation

algaGPT was introduced in:

@article{nelson2025la4sr,
  title   = {Pan-microalgal dark proteome mapping via interpretable deep learning and synthetic chimeras},
  author  = {Nelson, David R. and Jaiswal, Ashish Kumar and Ismail, Noha Samir and Mystikou, Alexandra and Salehi-Ashtiani, Kourosh},
  journal = {Patterns},
  volume  = {6},
  pages   = {101373},
  year    = {2025},
  doi     = {10.1016/j.patter.2025.101373}
}

algaGPT was applied at scale in:

@article{nelson2026elfnet,
  title   = {Coupling of oceanographic state to the dark proteome: a foundation for genome-informed marine productivity modeling},
  author  = {Nelson, David Roy and Plouviez, Maxence and Daakour, Sarah and Jaiswal, Ashish and Fu, Weiqi and Amin, Shady A. and Salehi-Ashtiani, Kourosh},
  journal = {Forthcoming},
  year    = {2026}
}

Contact

Kourosh Salehi-Ashtiani -- ksa3@nyu.edu