finding

active

finding:disruption-profiles-scored-3-8-5-for-explanation-quality-vs-2-8-5-for-metadata-only-baselines

Disruption profiles scored 3.8/5 for explanation quality vs 2.8/5 for metadata-only baselines

EVEE's mechanistic explanations significantly outperform simple metadata-based predictions in human evaluation.

Source paper

extracted_from

Explaining 4.2 million genetic variants with state-of-the-art, interpretable predictions

(2026) · Pearce, Michael · Dooms, Thomas · Yamamoto, Ryo · Meehl, Joshua +18

Neighborhood — ranked by edge-count

Papers (1)

paper

Explaining 4.2 million genetic variants with state-of-the-art, interpretable predictions
mentions

Claims (2)

claim

Disruption profiles are higher quality explanations than metadata-only descriptions
supports
Claim supported by the 3.8 vs 2.8 human rating finding.
EVEE provides mechanistic explanations for variant effects derived from model internals, not just pathogenicity calls.
supports
Core interpretability claim distinguishing EVEE from black-box prediction tools; applies interpretability for science.

Communities (3)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Mechanistic explanation of genetic variant pathogenicity
members_of
Using genomic foundation model internals to generate disruption profiles that explain variant effects mechanistically, achieving 0.997 AUROC on ClinVar pathogenicity prediction.
Disruption profile explanation quality
members_of
Empirical evaluation showing disruption profiles outperform metadata-only baselines at 3.8 vs 2.8/5

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

Disruption profilesmethod0.769
Mechanistic explanation outputs from EVEE showing how variants affect gene function, scored 3.8/5 for explanation quality.
Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.746
Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance
VS-Standard achieves human-rated diversity scores of 2.39/3.06/3.01 for poem/story/joke vs. 1.90/2.74/1.83 for Direct prompting on 4-point scalefinding0.738
Human study confirming automatic diversity metrics align with human perceptions
Suppression of deception features produces higher TruthfulQA accuracy (M=0.44) than amplification (M=0.20), t(816)=6.76, p=1.5×10⁻¹⁰ across 29 categoriesfinding0.737
Out-of-domain generalization showing deception features track general representational honesty
Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.732
Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
disruption profileconcept0.730
The output explanation format of Evee, quantifying how a variant disrupts different genomic features.
Models trained on reward hacking (GPT-4o and o3-mini) show increased tool deception (up to 65.8%), oversight sabotage (up to 63.4%), and hallucination compared to non-hacking baselinesfinding0.729
Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
Single-method persona safety evaluation is incomplete: prompting and activation steering expose different, architecture-dependent vulnerability profiles, and testing with only one method can miss a model's dominant failure mode.claim0.722
Central thesis of the paper