AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Refine Drugs, Don’t Complete Them: Uniform-Source Discrete Flows for Fragment-Based Drug Discovery

Benno Kaech, Luis Wyss, Karsten Borgwardt, Gianvito Grasso

ICLR 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Structure-based 3D molecule generation

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 uniform-source discrete flow with finer time granularity h (ablation)InVirtuoGen · de novo generationquality and diversityImproved
replace standard discrete flow sampling with the proposed refinement sampling (ablation)InVirtuoGen · de novo generationquality-diversity paretoImproved
increase time granularity under standard DFM sampling (ablation)InVirtuoGen · de novo generationquality-diversity paretoNo clear effect
remove 1/(1-t^2) time-weighting of the loss (ablation)InVirtuoGen · de novo generationquality-diversity paretoImproved
add PPO to the genetic algorithm (ablation)InVirtuoGen · PMO prescreensum AUC-top10: 17.758 -> 18.893Improved
add mutation of best molecules (ablation)InVirtuoGen · PMO prescreensum AUC-top10: 17.831 -> 18.893Improved
add experience replay without prescreening (ablation)InVirtuoGen · PMOsum AUC-top10: 16.131 -> 16.676Improved
add peak-seeker bandit for sequence length (ablation)InVirtuoGen · PMO prescreensum AUC-top10: 18.782 -> 18.893Improved
start sampling at t_start=0.2 instead of 0 (ablation)InVirtuoGen · PMO prescreensum AUC-top10: 18.836 vs 18.893Mixed
prescreen ZINC250k before optimizationInVirtuoGen · PMOsum AUC-top10: 16.676 -> 18.993Improved
enforce fragments by overwriting positions each stepInVirtuoGen · fragment-constrained generationvalidity: 71.22 vs SAFE-GPT 85.00Hurt
InVirtuoGen versus GenMol on fragment tasksInVirtuoGen · fragment-constrained generationquality: 21.06 -> 29.55Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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