You can deploy this model and start earning money today!

melolmg/
ExpressML

Views

No views yet

ExpressML — Protein Expression Predictor

ESM-2 150M fine-tuned on PaxDB for predicting protein soluble expression in E. coli.

Performance

Metric	ExpressML	SoluProt
AUC	0.867	0.620
Accuracy	0.788	—

Quick Start

bash

pip install torch fair-esm
wget https://huggingface.co/melolmg/ExpressML/resolve/main/esm2_paxdb_exp.pt
python express-esm --seq MKKTAIAIAVALAGFATVAQAAEP --model esm2_paxdb_exp.pt

Model Architecture

ESM-2 150M (frozen) → mean pooling → MLP(1280→512→128→1) → Sigmoid

Training

Parameter	Value
Dataset	PaxDB 161,459 proteins
Positive ratio	55.93%
Optimizer	AdamW (lr=2e-5, wd=0.01)
Epochs	3 (best at epoch 2)
Batch size	4
Max sequence length	512

Output Scores

Grade	Range	Meaning
A	≥ 0.75	High expression
B	0.50–0.75	Expressible, optimize
C	0.25–0.50	Difficult
D	< 0.25	Very difficult

Citation

Trained on PaxDB (Wang et al., 2015) with ESM-2 (Lin et al., 2023).