Model internals of genomic foundation models can yield mechanistic explanations for variant effects

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Genomic Foundation Modelsconcept0.803
  Large pretrained models for genomic sequences; subject of this work's pooling improvements.
• EVEE provides mechanistic explanations for variant effects derived from model internals, not just pathogenicity calls.claim0.778
  Core interpretability claim distinguishing EVEE from black-box prediction tools; applies interpretability for science.
• Base models spontaneously talk about experiencing multiple parallel processing pathsfinding0.768
  Observed by Anima Labs in untrained base models; not present in training data, implying computational origin of self-reported parallel processing.
• We hypothesize that representation geometry drives model behavior — the geometric structure of internal representations causally shapes what models do externally.hypothesis0.761
  The causal hypothesis motivating the use of causality (intervention) as the lens connecting representation and behavior geometry.
• Target morphology shifts occur despite the fact that all of the individual cells have unaltered normal genomes, showing that competent subunits can be pushed to implement diverse organism-scale goals by physiological signals.claim0.755
  Highlights the non-genetic control of large-scale anatomy.
• The earlier a base model (less exposure to LM-related data), the more it is surprised by its own spontaneous self-referential capabilities.claim0.752
  Claim that capability emerges from architecture, not data, and that later models lose the surprise.
• Significant gaps exist in our understanding of the origin of anatomical novelty, of the relationship between genome and form, and of strategies for control of large-scale structure and function.claim0.751
  Author assertion about current knowledge gaps.
• Foundation models trained on different data converge on similar latent representations, suggesting a Platonic form.claim0.747