We just updated our manuscript "Understanding Language Model Scaling on Protein Fitness Prediction". Where we explained why larger pLMs don’t always perform better on mutation effect prediction. We extended beyond ESM2 to models like ESMC, ESM3, SaProt, and ESM-IF1.

#ProteinLM

⚠️ Caution when using ProteinGYM binary classification: some DMS_binarization_cutoff values appear inverted, some are totally unreasonable given the DMS_score shows a clear bimodal distribution.

How can we better understand pathogenic variants in intrinsically disordered regions (IDRs)? How do models such as AlphaMissense and ESM1b predict pathogenicity, when these regions typically exhibit lower genomic conservation than ordered regions? Read more:
doi.org/10.1101/2025...

Protein language model likelihood are better zero shot mutation effect predictions when they have perplexity 3-6 on the wildtype sequence.

www.biorxiv.org/content/10.1...

Why do large protein language models like ESM2-15B underperform compared to medium-sized ones like ESM2-650M in predicting mutation effects? 🤔

We dive into this issue in our new preprint—bringing insights into model scaling on mutation effect prediction. 🧬📉

We have updated our protein lanuage model trained on structure dynamics. Our new models show significant better zero-shot performance on mutation effects of designed and viral proteins compared to ESM2. check the new preprint here: www.biorxiv.org/content/10.1...

SeqDance: A Protein Language Model for Representing Protein Dynamic Properties https://www.biorxiv.org/content/10.1101/2024.10.11.617911v1