Biology Atlas

Map / Open register

question · intermediate · partly written

Does choosing a representation per query beat one fixed representation?

Per-task winners and concatenations are documented; a trained per-query chooser compared with a tuned fixed fingerprint on a scaffold split was not found.

What the sources show more

Of 28 pretrained embedding models on 25 datasets, all neural models except CheMeleon failed to outperform the ECFP count fingerprint; 12 fingerprints plus two simple baselines differed little in virtual screening. 12

Reported, not yet replicated

What is missing more

A selector, nearest view or learned, with the choice made inside nested validation and scored on scaffold or temporal splits. 31

Our inference

What would settle it (our inference) more

Five to ten public ADMET sets; selector among ECFP count, RDKit descriptors and one embedding against ECFP count with equal tuning, and against the oracle best-per-task. 31

Our inference

Observation

Many encodings of one molecule exist and benchmarks compare them, but the choice is made by the benchmark's author. 32

Reported, not yet replicated

What is known

Of 28 pretrained embedding models on 25 datasets, all neural models except CheMeleon failed to outperform the ECFP count fingerprint; 12 fingerprints plus two simple baselines differed little in virtual screening. 12

Reported, not yet replicated

What is not

A selector, nearest view or learned, with the choice made inside nested validation and scored on scaffold or temporal splits. 31

Our inference

What would settle it

Five to ten public ADMET sets; selector among ECFP count, RDKit descriptors and one embedding against ECFP count with equal tuning, and against the oracle best-per-task. 31

Our inference

Still open here

Not yet written: Hypotheses, What is missing, Who works on it.

Linked from

Sources

  1. Praski M, Adamczyk J, Czech W. Benchmarking Pretrained Molecular Embedding Models For Molecular Representation Learning. arXiv (2025). preprint, not peer reviewed; opened 2026-11-18 all records citing this
  2. Riniker S, Landrum GA. Open-source platform to benchmark fingerprints for ligand-based virtual screening. Journal of Cheminformatics (2013). paper; opened 2026-11-18 all records citing this
  3. Wu Z, Ramsundar B, et al.. MoleculeNet: a benchmark for molecular machine learning. Chemical Science (2018). paper; opened 2026-11-18 all records citing this
Where it sits on the map
History of this record (5)

This record has been changed 5 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.

Kind
amit
Register
open-question
Triage
needs-research
Status
partly written
Last updated
2026-11-19
Record
content/questions/does-choosing-a-representation-per-query-beat-one-fixed.json