Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

Conservation of Protein Domains Over Different Proteins02:26

Conservation of Protein Domains Over Different Proteins

11.3K
Protein domains are small structurally independent units that are part of a single amino acid chain.  Although these domains are often structurally independent, they may rely on synergistic effects to perform their functions as part of a larger protein. Protein domains may be conserved within the same organism, as well as across different organisms.
A limited set of protein domains often duplicate and recombine during evolution. These domains can be organized in different combinations to...
11.3K
Improving Translational Accuracy02:07

Improving Translational Accuracy

11.9K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
11.9K
Leaky Scanning02:28

Leaky Scanning

5.2K
During most eukaryotic translation processes, the small 40S ribosome subunit scans an mRNA from its 5' end until it encounters the first start AUG codon. The large 60S ribosomal subunit then joins the smaller one to initiate protein synthesis. The location of the translation initiation is largely determined by the nucleotides near the start codon as there may be multiple translation initiation sites present on the mRNA.  Marilyn Kozak discovered that the sequence RCCAUGG (where R...
5.2K
Protein Networks02:26

Protein Networks

2.4K
2.4K
Conservation of Protein Domains02:26

Conservation of Protein Domains

3.2K
3.2K
Protein-Protein Interfaces02:04

Protein-Protein Interfaces

3.8K
3.8K

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Promera: a unified model for biomolecular structure prediction, filtering, and design.

bioRxiv : the preprint server for biology·2026
Same author

SwitchCraft: A Programmatic Framework for Designing State-Switching Proteins.

ArXiv·2026
Same author

Microbiome-Directed Therapy for Fatigue in Quiescent Inflammatory Bowel Disease: A Randomized Placebo-Controlled Trial of Multistrain Probiotic Supplementation.

The American journal of gastroenterology·2026
Same author

Dango: Predicting higher-order genetic interactions.

Cell systems·2026
Same author

Evolutionary dynamics under phenotypic uncertainty.

bioRxiv : the preprint server for biology·2026
Same author

Profile of David Baker, Demis Hassabis, and John Jumper: 2024 Nobel laureates in chemistry.

Proceedings of the National Academy of Sciences of the United States of America·2026

Related Experiment Video

Updated: Sep 10, 2025

Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
09:47

Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches

Published on: December 15, 2023

1.3K

Sparse autoencoders uncover biologically interpretable features in protein language model representations.

Onkar Gujral1,2, Mihir Bafna2, Eric Alm3,4

  • 1Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139.

Proceedings of the National Academy of Sciences of the United States of America
|August 19, 2025
PubMed
Summary

Sparse autoencoders extract interpretable features from protein language models (PLMs) without supervision. These features reveal biological insights, enhancing the explainability and trust of AI in life sciences.

Keywords:
interpretabilityprotein language modelssparse autoencoderstranscoders

More Related Videos

Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions
06:50

Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions

Published on: January 26, 2024

2.0K
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

681

Related Experiment Videos

Last Updated: Sep 10, 2025

Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
09:47

Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches

Published on: December 15, 2023

1.3K
Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions
06:50

Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions

Published on: January 26, 2024

2.0K
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

681

Area of Science:

  • Computational Biology
  • Artificial Intelligence in Life Sciences
  • Bioinformatics

Background:

  • Protein language models (PLMs) have advanced biological predictions but suffer from a "black-box" nature, limiting transparency and explainability.
  • Interpreting the features learned by PLMs is crucial for human-AI collaboration and understanding biological mechanisms.
  • Existing methods for feature interpretation often require supervision, which can be labor-intensive and may not capture all relevant information.

Purpose of the Study:

  • To develop an unsupervised method for extracting interpretable features from protein-level and amino acid-level representations of PLMs.
  • To assess the biological relevance and interpretability of the extracted sparse features.
  • To enhance the safety, trust, and explainability of PLMs in biological applications.

Main Methods:

  • Leveraged sparse autoencoders (SAEs) and transcoders, adapted from natural language processing, to extract features from the ESM2 PLM.
  • Performed feature extraction in a completely unsupervised manner, without relying on external biological labels or probes.
  • Utilized Anthropic's Claude AI to automate the interpretation of extracted sparse features.

Main Results:

  • SAEs successfully extracted interpretable sparse features from both protein-level and amino acid-level representations of ESM2.
  • Many extracted sparse features showed strong associations with Gene Ontology (GO) terms across all hierarchy levels.
  • Automated interpretation identified features corresponding to specific protein families (e.g., NAD Kinase, PTH), functions (e.g., methyltransferase activity), and sensory perceptions.

Conclusions:

  • SAEs provide a powerful unsupervised approach for disentangling biologically relevant information within PLM representations.
  • The extracted sparse features are more interpretable than raw PLM neurons, aiding biological insight discovery.
  • This work significantly advances the explainability and trustworthiness of PLMs, paving the way for deeper biological understanding.