AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Effects of Structural Reward Shaping on Biophysical Properties in RL-Trained Plasmid Generators

McClain Thiel, Angus G. Cunningham, Chris P Barnes

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 1 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
add the cassette arrangement bonus (promoter-CDS-terminator order) to the reward (ablation)GRPO-trained PlasmidGPT · 8 prompts, T=0.95QC pass rate: 19.8% -> 66.9%Improved
add GRPO structural-reward post-training to pretrained PlasmidGPTPlasmidGPT · 8 prompts, 4,000 sequencesQC pass rate: 4.3% -> 71.6%Improved
supervised fine-tuning on curated plasmids instead of RLPlasmidGPT · 8 prompts, 4,000 sequencesQC pass rate: 4.3% -> 11.0%Improved
rejection sampling more heavily from the base model (K=64) (ablation)Base PlasmidGPT · Rejection sampling successsuccess rate: 4.3% -> 54.5% (K=1 to K=64), still far below RL 76.8% at K=1No clear effect
RL post-training reduces output diversityPlasmidGPT · pooled 21-mer Jaccard distancediversity: 0.960 -> 0.596Hurt
RL post-training on held-out plasmid continuation vs BasePlasmidGPT · 29 held-out non-Addgene plasmidsmean log-probability: -8.99 -> -8.16Improved
RL instead of SFT on held-out continuation log-probabilityPlasmidGPT · 29 held-out non-Addgene plasmidsmean log-probability: -7.35 -> -8.16Hurt
add structural reward post-training (SFT or RL) on MFE densityPlasmidGPT · Generated sequencesMFE density (kcal/mol/bp): -0.105 -> -0.149 (real reference -0.151)Improved
use only a single reward component (length-only or CDS-only) (ablation)GRPO reward · MFE density of generated sequencesMFE density (kcal/mol/bp): -0.149 (full) vs -0.126 (length-only), -0.103 (CDS-only)Hurt
RL reward shaping on GC contentPlasmidGPT · Generated sequencesGC content: RL 0.524 vs real 0.510Mixed

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