Research map / Learning theory and algorithms / Submodular and game algorithms
Constrained Dynamic Programming and Supervised Penalty Learning Algorithms for Peak Detection in Genomic Data
Toby Dylan Hocking, Guillem Rigaill, Paul Fearnhead, Guillaume Bourque
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add the up-down constraint on adjacent segment means (GPDPA) instead of unconstrained PDPA (ablation) | changepoint peak detection · ChIP-seq benchmark, test AUC | test AUC: PDPA significantly lower test AUC in all seven data sets (p-value < 0.04) | Improved |
| use exact GPDPA instead of heuristic CDPA | up-down constrained peak detection · ChIP-seq benchmark, test AUC | test AUC: 0.873 vs 0.887 (p=0.056); 0.913 vs 0.897 (p=0.02) | No clear effect |
| use GPDPA instead of CDPA for speed | up-down constrained peak detection · all 2752 ChIP-seq vectors | total computation time: 156 hours -> 6 hours (26 times faster) | Improved |
| use GPDPA instead of MACS and HMCanBroad heuristics | peak detection · ChIP-seq benchmark, label error | minimum incorrect labels: GPDPA better for 1219 problems vs HMCanBroad better for 39 | Improved |
| use Remove rule instead of Join rule for active equality constraints (ablation) | GPDPA · ChIP-seq benchmark, label error | minimum incorrect labels: Remove better for 51 problems, Join better for 18 problems | Improved |
| use oracle model complexity instead of linear model complexity (ablation) | penalty learning with CDPA, GPDPA, PDPA · ChIP-seq benchmark, test AUC | test AUC: mean difference from 0.00004 to 0.0006, p-value > 0.05 | No clear effect |
| learn multi-parameter penalty function instead of single-parameter (ablation) | changepoint penalty learning · ChIP-seq benchmark, test AUC | test AUC: mean difference from 0.003 to 0.005, p-value < 0.07 | Improved |
| supervised penalty learning instead of unsupervised default thresholds (ablation) | peak detection algorithms · ChIP-seq benchmark, test accuracy | percent correctly predicted labels: significantly higher for almost every algorithm and data set; exception H3K36me3 TDH other, 6.9-8.2% accuracy difference | Improved |
| multi-parameter penalty instead of single-parameter, percent accuracy (ablation) | GPDPA · ChIP-seq benchmark, test accuracy | percent correctly predicted labels: mean difference of 1-3% accuracy, p-value < 0.05 | Improved |
Closest papers by meaning
- SPACE: Your Genomic Profile Predictor is a Powerful DNA Foundation Model (ICML 2025 · similarity 0.82)
- Hyperbolic Genome Embeddings (ICLR 2025 · similarity 0.82)
- GLACIER: Rethinking Mass Spectrum Prediction as an Object Detection Problem (NeurIPS 2026 · similarity 0.81)
- Optimal Classification Trees for Continuous Feature Data Using Dynamic Programming with Branch-and-Bound (AAAI 2025 · similarity 0.81)
- BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks (ICLR 2024 · similarity 0.81)
- Context-Aware Regularization with Markovian Integration for Attention-Based Nucleotide Analysis (NeurIPS 2025 · similarity 0.81)
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.
