Paper List
-
Emergent Spatiotemporal Dynamics in Large-Scale Brain Networks with Next Generation Neural Mass Models
This work addresses the core challenge of understanding how complex, brain-wide spatiotemporal patterns emerge from the interaction of biophysically d...
-
Human-Centred Evaluation of Text-to-Image Generation Models for Self-expression of Mental Distress: A Dataset Based on GPT-4o
This paper addresses the critical gap in evaluating how AI-generated images can effectively support cross-cultural mental distress communication, part...
-
GOPHER: Optimization-based Phenotype Randomization for Genome-Wide Association Studies with Differential Privacy
This paper addresses the core challenge of balancing rigorous privacy protection with data utility when releasing full GWAS summary statistics, overco...
-
Real-time Cricket Sorting By Sex A low-cost embedded solution using YOLOv8 and Raspberry Pi
This paper addresses the critical bottleneck in industrial insect farming: the lack of automated, real-time sex sorting systems for Acheta domesticus ...
-
Collective adsorption of pheromones at the water-air interface
This paper addresses the core challenge of understanding how amphiphilic pheromones, previously assumed to be transported in the gas phase, can be sta...
-
pHapCompass: Probabilistic Assembly and Uncertainty Quantification of Polyploid Haplotype Phase
This paper addresses the core challenge of accurately assembling polyploid haplotypes from sequencing data, where read assignment ambiguity and an exp...
-
Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors
This paper addresses the core challenge of automating the discovery of biologically plausible recurrent neural network (RNN) dynamics that can replica...
-
Influence of Object Affordance on Action Language Understanding: Evidence from Dynamic Causal Modeling Analysis
This study addresses the core challenge of moving beyond correlational evidence to establish the *causal direction* and *temporal dynamics* of how obj...
Mechanistic Interpretability of Antibody Language Models Using SAEs
Department of Statistics, University of Oxford, UK | Reticular, San Francisco, USA | Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), Cambridge, MA, USA
The 30-Second View
IN SHORT: This work addresses the core challenge of achieving both interpretability and controllable generation in domain-specific protein language models, specifically for antibody design.
Innovation (TL;DR)
- Methodology First application of Sparse Autoencoders (SAEs) to interrogate autoregressive antibody-specific language models (p-IgGen), moving beyond general protein language models.
- Methodology Systematic comparison reveals a key trade-off: TopK SAEs yield highly interpretable, monosemantic features (e.g., for CDR identity with validation accuracy 0.99) but lack causal steerability, while Ordered SAEs provide reliable generative control at the cost of interpretability.
- Biology Identifies and validates antibody-specific, biologically meaningful latent features, such as CDR identity and germline gene identity (e.g., IGHJ4 prediction with F1 macro score of 0.93), demonstrating the model's learning of immunologically relevant concepts.
Key conclusions
- TopK SAEs effectively compress and preserve biological information (CDR identity prediction accuracy 0.99 vs. 0.98 for raw neurons) and yield sparse, interpretable activation patterns localized to specific regions (e.g., CDRH3), overcoming neuron polysemanticity.
- High feature-concept correlation (e.g., F1 > 0.5 for IGHJ4 latents) does not guarantee causal steerability; steering on TopK-identified IGHJ4 features failed to consistently increase IGHJ4 proportions in generated sequences.
- Ordered SAEs, with their enforced hierarchical latent structure (via per-index nested grouping and decreasing truncation weights), successfully identify features that enable predictable generative steering, albeit with more complex activation patterns.
Abstract: Sparse autoencoders (SAEs) are a mechanistic interpretability technique that have been used to provide insight into learned concepts within large protein language models. Here, we employ TopK and Ordered SAEs to investigate an autoregressive antibody language model, p-IgGen, and steer its generation. We show that TopK SAEs can reveal biologically meaningful latent features, but high feature–concept correlation does not guarantee causal control over generation. In contrast, Ordered SAEs impose an hierarchical structure that reliably identifies steerable features, but at the expense of more complex and less interpretable activation patterns. These findings advance the mecahnistic interpretability of domain-specific protein language models and suggest that, while TopK SAEs suffice for mapping latent features to concepts, Ordered SAEs are preferable when precise generative steering is required.