Related Experiment Video
Updated: Jun 17, 2025
06:50
Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions
Published on: January 26, 2024
1.8K
Leveraging a large language model to predict protein phase transition: A physical, multiscale, and interpretable
Mor Frank1,2, Pengyu Ni1,2, Matthew Jensen1,2
1Program in Computational Biology and Bioinformatics, Yale University, New Haven, CT 06520.
Summary
Large language models (LLMs) can now predict protein phase transitions (PPTs), distinguishing between liquid droplets and solid aggregates. This unified approach aids in understanding disease mechanisms and protein design, outperforming older methods.
Area of Science:
- Biochemistry
- Computational Biology
- Neuroscience
Background:
- Protein phase transitions (PPTs), including liquid-liquid phase separation and amyloid aggregation, are implicated in age-related diseases like Alzheimer's.
- Existing computational tools predict droplet or aggregate formation separately, lacking a unified framework.
- Large language models (LLMs) show promise in protein structure prediction but haven't been applied to PPTs.
Purpose of the Study:
- To develop and validate a unified computational framework using LLMs for predicting both liquid and solid protein phase transitions.
- To assess the impact of protein sequence variants on PPTs for applications in protein design.
- To compare the performance of the LLM-based framework against classical computational benchmarks.
Main Methods:
- Fine-tuning a large language model (LLM) for the prediction of protein phase transitions (PPTs).
- Utilizing a random forest model with biophysical features to interpret LLM predictions.
- Analyzing Alzheimer's disease-related proteins to investigate the relationship between aggregation propensity and gene expression.
Main Results:
- The fine-tuned LLM successfully predicts both liquid and solid protein phase transitions within a unified framework.
- The LLM-based approach demonstrates superior performance compared to established computational benchmarks.
- Interpretation using a random forest model provides insights into the biophysical drivers of PPTs.
- Reduced gene expression of aggregation-prone proteins was observed in Alzheimer's disease.
Conclusions:
- LLMs offer a powerful, unified approach for predicting diverse protein phase transitions (PPTs).
- This method facilitates the study of sequence variant effects on PPTs, aiding protein design and disease research.
- Findings suggest a potential natural defense mechanism against protein aggregation in Alzheimer's disease.
