whiteh4t/esm2-35m-protein-localization

ESM-2 35M — Protein Subcellular Localization (Full Fine-Tune)

Fine-tuned ESM-2 35M for protein subcellular localization (10 classes). All parameters trained end-to-end.

Results

Metric	Score
Accuracy	74.3%
F1 (macro)	0.647
F1 (weighted)	0.737
MCC	0.677

Architecture

ESM-2 35M (all trainable) → Mean pooling → LayerNorm → Linear(480→120) → GELU → Linear(120→10)

Training

• Dataset: DeepLoc 2.0 (17,266 train / 3,700 val / 3,701 test)
• Strategy: Full fine-tune (all 35M parameters)
• Epochs: 10
• Learning rate: 2e-5
• Hardware: NVIDIA DGX Spark

All Models in This Project

Model	Strategy	Accuracy	Link
ESM-2 8M	Linear probe	69.6%	whiteh4t/esm2-8m-protein-localization
ESM-2 35M	Full fine-tune	74.3%	This repo
ESM-2 150M	Full fine-tune	76.6%	whiteh4t/esm2-150m-protein-localization
ESM-2 650M	LoRA (r=16)	76.5%	whiteh4t/esm2-650m-protein-localization-lora

Links

• Demo: HuggingFace Space
• Code: GitHub