Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23572957Tasks and data. 17 tasks (ChEMBL target × Ki or IC50) extracted from ChEMBL release 37 via the REST API on 10 October 2026, restricted to binding assays with confidence score ≥ 8, relation “=”, a pChEMBL value and no validity flags. Structures are standardised with RDKit. For every task: a molecule file (identifier, standardised SMILES, label averaged over documents, number of source documents, year of first report) and the underlying per-document measurements. Both the decontaminated set (molecules occurring in, or near-analogous to, MoleculeNet and TDC removed) and the uncleaned set are included. Noise floors. Per-task estimates of the between-document standard deviation, the implied RMSE floor and maximum achievable R², with bootstrap intervals over molecules and over document pairs, and sensitivity analyses for trimming, duplicated values, stereochemistry and assay selection. Benchmark results. Predictions, metrics and summary tables for ECFP fingerprints with gradient boosting, random forest and MLP; frozen ChemBERTa-77M and MoLFormer-XL embeddings with a boosted-tree head; and, on a five-task subset, a D-MPNN and a fine-tuned ChemBERTa. All under random, scaffold and temporal splits with five seeds, plus minimum detectable effects and slice analyses. Code and paper. The full pipeline (extraction, standardisation, decontamination, ceiling estimation, benchmarking, analysis) and the LaTeX sources and figures of the preprint. The README explains how to regenerate every table and figure from the stored results in minutes, or to rerun the models in about one and a half days on a laptop.
Keyword(s):
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
No Associated Publication Data Available
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)