AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Conditionally Site-Independent Neural Evolution of Antibody Sequences

Stephen Zhewen Lu, Aakarsh Vermani, Kohei Sanno, Jiarui Lu, Frederick A Matsen, Milind Jagota, Yun S. Song

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Protein fitness and variant effect prediction · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
subtract Thrifty SHM likelihood to form a selection score instead of raw log-likelihood (ablation)CoSiNE · FLAb2 DMS assays (7 datasets)Spearman correlation: e.g. Koenig expression heavy 0.53 -> 0.61Improved
input only the heavy or light chain instead of the paired antibody (ablation)CoSiNE selection score · FLAb2 DMS assaysSpearman correlationHurt
vary branch length t from 0.1 to 0.4 in the selection score (ablation)CoSiNE · FLAb2 DMS assaysSpearman correlation: delta rho <= 0.045No clear effect
train from scratch without the ESM2 backbone (ablation)CoSiNE · FLAb2 DMS assaysSpearman correlation: -0.041 on averageHurt
condition rates on sequence context (CoSiNE) instead of a WAG-style context-independent matrix (ablation)CoSiNE · antibody test transitionsper-token NLLImproved
use CoSiNE instead of DASM+Thrifty for transition likelihoodCoSiNE · Rodriguez et al. test transitionsper-site likelihood: better in 62.3% of casesImproved
use CoSiNE instead of ProGen2-Small on Koenig light-chain expressionCoSiNE · FLAb2 Koenig expression (light chain)Spearman correlation: 0.508 vs 0.513Hurt
use CoSiNE instead of best baselines on Adams (mouse wildtype)CoSiNE · FLAb2 Adams expressionSpearman correlation: 0.297 -> 0.464Improved
add classifier guidance (Guided Gillespie) with gamma from 1 to 15 (ablation)CoSiNE sampling · CoV-AbDab SARS-CoV-1 affinity maturationpredicted affinity gainImproved
set guidance strength gamma >= 10 (ablation)Guided Gillespie · CoV-AbDab SARS-CoV-1 affinity maturationpredicted affinity gainMixed
add guidance (gamma = 5) to sampling (ablation)CoSiNE · AbodyBuilder3 pLDDT and OASisstructural plausibility and humannessNo clear effect
use Taylor-approximated guidance instead of exact oracle guidance (ablation)Guided Gillespie · antibody guidancespeed and fitness improvement: 500-fold speedup, no significant fitness differenceImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy