AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Representing local protein environments with machine learning force fields

Meital Bojan, Sanketh Vedula, Sai Advaith Maddipatla, Nadav Bojan, Anar Rzayev, Federico Napoli, Paul Schanda, Alexander Bronstein

ICLR 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Protein interface and docking prediction

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
use MLFF embeddings (AIMNet) instead of ESM-based or learned featuresGCN pKa predictor · pKa, Glutamic acidMAE: 0.518 (learned) -> 0.265Improved
use MACE embeddings instead of UCBShift for CAchemical shift predictor · backbone CAMAE (ppm): 0.653 -> 0.584Improved
use MACE embeddings instead of UCBShift for HAchemical shift predictor · backbone HAMAE (ppm): 0.165 -> 0.180Hurt
use 5 A radius instead of 4 A or 6 A for the environment (ablation)MACE + GCN · backbone chemical shiftsMAE (ppm): CA 0.620 / 0.627 -> 0.584Improved
concatenate first and second MACE message-passing layers (L1+2) (ablation)MACE + GCN · backbone N shiftMAE (ppm): 1.771 -> 1.642Improved
use deeper Orb layer (L15 instead of L5/L10) (ablation)Orb + GCN · backbone N shiftMAE (ppm): 1.889 -> 2.015Hurt
use six backbone atoms instead of CA only for secondary structure (ablation)Egret + GCN · secondary structureweighted F1 (%): 94.817 -> 98.633Improved
use six backbone atoms instead of CA only for amino acid identity (ablation)Egret + GCN · amino acid identityweighted F1 (%): 95.629 -> 99.135Improved
use MLFF embeddings instead of hand-crafted LocoHD descriptorsGCN · secondary structureweighted F1 (%): 76.323 -> 95.693Improved
use kernel-density likelihood in MLFF embedding space as confidenceMACE shift predictor · chemical shift CA and Nprediction errorImproved
use MACE instead of pKa-ANI for histidine pKaGCN · PKAD-R histidineMAE: 0.899 -> 0.925Hurt
use AIMNet embeddings for pKa vs MACE embeddingsGCN · pKa predictionMAE: Lysine 0.320 (MACE) vs 0.270 (AIMNet)Mixed

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