Research map / Probabilistic and scientific modeling / Molecules and proteins
KinDEL: DNA-Encoded Library Dataset for Kinase Inhibitors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces KinDEL, an 81M-compound DNA-encoded library dataset screened against MAPK14 and DDR1 kinases with on- and off-DNA biophysical validation, and benchmarks ML models including the probabilistic DEL-Compose.
Problem
Publicly available DNA-encoded library (DEL) datasets are scarce, and DEL data is noisy and biased by the DNA tag, so ML methods for hit identification lack benchmarks that validate predictions against off-DNA affinity.
Approach
Release a large trisynthon DEL (~81M molecules) against two kinases with docked 3D poses and FP (on-DNA) and SPR (off-DNA) K_D validation data, then benchmark regression baselines against probabilistic count-based models (DEL-Compose) under random, disynthon, and cluster splits.
Key results
- On MAPK14 off-DNA in-library, DEL-Compose (synthon) reached Spearman 0.760 vs 0.353 for count enrichment and 0.166 for Poisson enrichment.
- On DDR1, DEL-Compose (full molecule) was the best in-library off-DNA model (0.509 Spearman) and DEL-Compose generally outperformed enrichment baselines off-DNA.
- Almost all models performed worse on the disynthon split than random split, most markedly for MAPK14.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DEL-Compose instead of raw count enrichment for off-DNA ranking | MAPK14 · In-Library off-DNA, random split | Spearman's ρ: 0.353 -> 0.760 ± 0.018 | Helped |
| Use DEL-Compose instead of Poisson enrichment for off-DNA ranking | DDR1 · In-Library off-DNA, random split | Spearman's ρ: 0.441 -> 0.509 ± 0.024 | Helped |
| Evaluate on a disynthon split rather than a random split | MAPK14 RF · In-Library on-DNA | Spearman's ρ: 0.694 ± 0.030 -> 0.157 ± 0.138 | Hurt |
| Evaluate on a disynthon split rather than a random split | DDR1 DEL-Compose (M) · In-Library on-DNA | Spearman's ρ: 0.731 ± 0.016 -> 0.663 ± 0.022 | Hurt |
Surprising findings
- Simple models (RF, kNN) performed best for MAPK14 in-library on-DNA, while DEL-Compose was better off-DNA.
- DEL-Compose ranked off-DNA affinities better than the raw DEL enrichment metrics themselves, suggesting a denoising effect.
Limitations
- DEL data only measures on-DNA binding, which correlates only loosely with off-DNA binding.
- Validation set is small (30-50 resynthesized molecules per target) and the extended set is limited in size.
- No counter-screen with a known inhibitor to distinguish allosteric from orthosteric binders.
Experimental setup
Datasets: KinDEL
Benchmarks: KinDEL MAPK14 held-out on-DNA and off-DNA K_D KinDEL DDR1 held-out on-DNA and off-DNA K_D
Metrics: Spearman correlation test loss
Baselines: Counts enrichment Poisson enrichment Random forest XGBoost kNN DNN GIN Chemprop
Builds on: DEL-Compose Poisson enrichment Chemprop
Closest papers by meaning
- WelQrate: Defining the Gold Standard in Small Molecule Drug Discovery Benchmarking (NeurIPS 2024 · similarity 0.84)
- S-MolSearch: 3D Semi-supervised Contrastive Learning for Bioactive Molecule Search (NeurIPS 2024 · similarity 0.84)
- Mol-Instructions: A Large-Scale Biomolecular Instruction Dataset for Large Language Models (ICLR 2024 · similarity 0.82)
- From Artificially Real to Real: Leveraging Pseudo Data from Large Language Models for Low-Resource Molecule Discovery (AAAI 2024 · similarity 0.82)
- Towards precision protein-ligand affinity prediction benchmark: A Complete and Modification-Aware DAVIS Dataset (NeurIPS 2025 · similarity 0.82)
- Offline Model-based Optimization for Real-World Molecular Discovery (ICML 2025 · similarity 0.82)
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.