Infoblock alpha
q-bio/biomolecules IB-arxiv-2608.18604-bll-top-70pct-of-27 measured

Across 27 curated multiparameter protein datasets and 2,592 models (three Bayesian architectures × 16 sequence representations, with dimensionality reduction), Bayesian Last Layer models ranked top-performing on 70% of the datasets.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

n_datasets 27 · n_models 2592 · bll_top_share_pct 70

01

Primary source

  1. arxiv · 2608.18604 · v1 · abstract
Title
Multitask Bayesian Neural Networks for Multiparameter Protein Engineering
Authors
Fabio Herrera-Rocha, David Medina-Ortiz, Desiree Wyrzykala, Tharun Srinivasan Sudha, Mehdi D. Davari
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
Bayesian Last Layer models delivered the strongest overall accuracy, generalization, and calibration, ranking as the top-performing model on 70% of benchmark datasets

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.