TmProt

Protein Thermostability Predictor

TmProt is a machine-learning-based protein thermostability predictor that leverages a fine-tuned ESM-2 protein language model to estimate melting temperatures (Tm) of protein sequences. It enables users to upload protein sequences in FASTA format (either pasted as text or uploaded as a file), and outputs predicted Tm values ranked by a user-defined thermostability threshold.

Paper: https://doi.org/10.64898/2026.05.07.723192
GitHub: https://github.com/loschmidt/TmProt
Select thermostability threshold (°C)

Default is 60°C

Results

Click an example to try TmProt instantly
FASTA sequences Select thermostability threshold (°C)

Features

- Predict protein melting temperature (Tm) from amino acid sequences

- Accepts input via FASTA text or FASTA file upload

- Supports sequences from 20 to 2000 amino acids in length

- Outputs a ranked table with predicted Tm and thermostability status based on user-chosen threshold

- CSV download option for easy export and downstream analysis

Model Overview

- Base Model: facebook/esm2_t33_650M_UR50D (650M parameters)

- Fine-tuning method: LoRA (Low-Rank Adaptation) using PEFT framework

- Task: Regression prediction of protein melting temperature (Tm)

- Training Data: ProMelt dataset (merged Meltome Atlas + ProTherm) with ~45,000 protein sequences and experimental Tm values

- Output: Single linear regression output neuron predicting Tm in °C