TARA-XGBoost-Bidirectional

Bidirectional XGBoost ensemble models linking marine environmental variables to microalgal protein domain (Pfam) abundance profiles from the TARA Oceans metagenomic dataset. These are the primary predictive models in the ELF-NET study.

Model Description

Forward Models (Environment → Pfam)

Property	Value
Input	32 Google Earth Engine oceanographic variables
Output	CLR-transformed abundance of 100 top-variance Pfam domains
Architecture	100 independent XGBoost regressors (one per target domain)
Performance	Median test R² = 0.20 (IQR: 0.16--0.29; max R² = 0.54)

Reverse Models (Pfam → Environment)

Property	Value
Input	9,611 Pfam domain abundances (100 PCA components, 72.2% variance)
Output	31 environmental variables
Architecture	31 independent XGBoost regressors (one per target variable)

Key results under 10-fold spatial block CV:

Target	Spatial Block CV R²	80/20 Split R²
Sea surface temperature	0.38	0.61
Bathymetry	0.42	0.53

Cross-basin validation (bathymetry): R² = 0.25.

Hyperparameters

Parameter	Value
n_estimators	200
max_depth	6
learning_rate	0.1
subsample	0.8
colsample_bytree	0.8
min_child_weight	3
reg_alpha	0.1
reg_lambda	1.0

Files

File	Size	Description
xgboost_forward_models_20260124_104452.joblib	49 MB	100 forward models (env → Pfam)
xgboost_reverse_models_20260124_104452.joblib	15 MB	31 reverse models (Pfam → env)
model_manifest_20260124_104452.json	--	Feature lists and hyperparameters

Usage

import joblib

# Load reverse models (Pfam → Environment)
reverse_bundle = joblib.load("xgboost_reverse_models_20260124_104452.joblib")

# Load forward models (Environment → Pfam)
forward_bundle = joblib.load("xgboost_forward_models_20260124_104452.joblib")

Dataset

Property	Value
Source	algaGPT-extracted proteomes from 2,044 TARA Oceans metagenome assemblies
Samples	1,279 (SMART-filtered) to 1,810 (GPS-recovered with WOA23 nutrients)
Environmental data	Google Earth Engine satellite products + WOA23 nutrients

Related Resources

Resource	Link
ELF-NET analysis pipeline (371 scripts, 15 modules)	github.com/olympus-terminal/ELF-NET
algaGPT protein classifier	GreenGenomicsLab/algaGPT
VICReg joint embedding model	TARA-WorldModel-VICReg
Dark-whiteGPLM checkpoints	SarahDaakour/dark-whiteGPLM
Domain-environment association data (Data S1)	Zenodo 10.5281/zenodo.18538439

Authors

David R. Nelson, Kourosh Salehi-Ashtiani

New York University Abu Dhabi

Citation

@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