concept · intermediate · partly written
"Quantum fingerprint": the metaphor, quantum-chemistry descriptors and quantum hardware, kept apart
Three things share the word: a metaphor, descriptors from approximate electronic structure on ordinary computers, and quantum circuits on hardware; none beat ordinary fingerprints on an activity task.
His idea, in his words
"conditional graph connections or entanglements, similar to quantum"; "It would be cool to invent a quantum fingerprint of some kind." 1
Amit's hypothesis
Thing one: the metaphor more
As a description of context-dependence it adds nothing beyond conditional dependence (see the record on context-dependent edges). To be more it would have to predict something classical conditional probability cannot, for example a Bell-type violation in how annotations co-vary; no source opened claims that for biological data. Where entanglement was removed from 12 quantum models on 160 simulated datasets, performance often stayed as good or better, and out-of-the-box classical models generally won. Quantum kernels, the usual route to a quantum fingerprint similarity, can concentrate exponentially as qubits grow, with entanglement one of four named causes. 23
Reported, not yet replicated
Thing two: quantum-chemistry descriptors on classical computers more
A classical machine solves an approximate electronic-structure problem for each molecule and charges, orbital energies or polarisabilities become features; this is not quantum computing. Rupp and colleagues predicted atomisation energies of over 7,000 molecules from such inputs, and Faber and colleagues predicted 13 QM9 properties; those targets are themselves computed quantum quantities, so quantum inputs are favoured by construction. Quantum-informed atom pairs gave a 0.5 to 6% mean R-squared gain over plain atom pairs on 10 datasets under repeated 10-fold cross-validation, with a slight loss on BACE-1, and RDKit descriptors ranked best overall. No head-to-head of such descriptors against ECFP on one dataset and one scaffold split was found. A 2026 study of quantum-mechanical descriptors was not readable at source check, so its figures are not used. 4567
Reported, not yet replicated
Thing three: quantum hardware more
Variational solvers have run on hardware for molecules up to BeH2 with up to six qubits. The one biology-adjacent hardware run found: a 16-qubit circuit samples a prior for a classical generator of KRAS (KRAS proto-oncogene, GTPase) inhibitors. It produced 21.5% more molecules passing synthesisability and stability filters than a vanilla LSTM, 15 molecules were synthesised, and one bound KRAS-G12D at 1.4 uM by surface plasmon resonance; on two of three Tartarus benchmark targets the circuit scored below the best classical algorithms, and both arms were carried forward, so the 15 do not isolate a quantum effect. It generates a distribution; it is not a fingerprint. 8910
Reported, not yet replicated
What remains open (our inference) more
A quantum-derived representation of a drug-like molecule beating ECFP on a scaffold split, on hardware or in simulation: none found. A testable non-classical prediction for annotation data that the metaphor would need to carry: none found. 62
Our inference
Cheapest test (our inference) more
Add semi-empirical charges to an ECFP count on one public ADMET set and compare with ECFP count alone, on a scaffold split with matched tuning. It runs in CPU-days with no quantum computer and tests the descriptor, not the metaphor. 64
Our inference
Still open here
- Does a quantum-derived representation add anything to ECFP on a biological endpoint under a scaffold split?
Not yet written: Plain answer, Why it matters, Equation, Real numbers.
Linked from
Sources
- Amit's dictation of 2026-11-18: how biology data is represented, and where ontologies can be innovated. Kept in the atlas repository; not published.
- Bowles J, Ahmed S, Schuld M. Better than classical? The subtle art of benchmarking quantum machine learning models. arXiv (2024). preprint, not peer reviewed; opened 2026-11-18 all records citing this
- Thanasilp S, Wang S, Cerezo M, et al.. Exponential concentration in quantum kernel methods. arXiv (2022). preprint, not peer reviewed; opened 2026-11-18 all records citing this
- Rupp M, Tkatchenko A, Müller KR, von Lilienfeld OA. Fast and accurate modeling of molecular atomization energies with machine learning. Physical Review Letters (2012). paper; opened 2026-11-18 all records citing this
- Faber FA, Hutchison L, Huang B, et al.. Machine learning prediction errors better than DFT accuracy. arXiv (2017). preprint, not peer reviewed; opened 2026-11-18 all records citing this
- Fliszkiewicz B, Sajdak M. Toward Quantum-Informed Atom Pairs. ACS Omega (2024). paper; opened 2026-11-18 all records citing this
- Hinostroza Caldas A, Kokorin A, Tkatchenko A, Medrano Sandonas L. Assessing the performance of quantum-mechanical descriptors in physicochemical and biological property prediction. Digital Discovery (2026). paper; opened 2026-11-18 all records citing this
- Peruzzo A, McClean J, et al., Aspuru-Guzik A, O'Brien JL. A variational eigenvalue solver on a photonic quantum processor. Nature Communications (2014). paper; opened 2026-11-18 all records citing this
- Kandala A, Mezzacapo A, et al.. Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets. Nature (2017). paper; opened 2026-11-18 all records citing this
- Ghazi Vakili M, Gorgulla C, et al., Aspuru-Guzik A. Quantum-computing-enhanced algorithm unveils potential KRAS inhibitors. Nature Biotechnology (2025). 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.
- Status
- partly written
- Last updated
- 2026-11-19
- Record
- content/concepts/quantum-fingerprint-and-the-entanglement-analogy.json