What Makes a Molecule Beautiful? Quantifying Collective Aesthetic Judgments of Molecular Structure
10,513

About Molesthetics

Molesthetics is a public experiment on molecular aesthetics. You are shown two standardized 2D structures and asked which one looks more appealing. There are no right answers — only collective visual preferences.

So far, 10,513 votes have been recorded from anonymous participants. Each click helps map how humans respond to molecular form when chemistry is hidden behind the drawing alone.

What you see

  • Blind pairwise comparisons (no names, scores, or IDs)
  • Identical rendering style for every structure
  • Random rotations to limit orientation effects

Molecular set

The platform uses 19,741 natural-product structures from the LOTUS database, drawn with RDKit under fixed graphical settings.

How pairs are chosen

Molecule A is preferentially picked among under-sampled structures. Molecule B is chosen either at random (~30%) or among structures with a similar current rating (~70%), balancing coverage of chemical space with local comparisons.

Privacy

Participation is anonymous. We store a random session cookie, your choice, and technical metadata (timestamp, rotations, trap flags). We do not collect names, e-mail addresses, or demographics.

Live scoring

After each vote, molecule ratings are updated online with an Elo system — a fast approximation to pairwise ranking. The map colors structures by these live Elo ratings.

Research analysis

Quality checks

A small fraction of duels are control comparisons (“traps”) with the same molecule on both sides, used to monitor positional bias. Regular duels show no strong left/right preference in aggregate.

Limitations

Judgments apply to 2D depictions, not 3D conformations. Most molecules receive only one vote, so rankings are exploratory rather than definitive. Open online participation means we do not know participants’ chemistry background.

Technical details (Elo update)

Expected score for molecule A against B:

P(A > B) = 1 / (1 + 10(RB − RA) / 400)

Rating update after a comparison:

R′ = R + K(S − E),  K = max(Kmin, K0 / √(n + 1))

References. (1) Rutz, A. et al. eLife 2022, 11, e70780 (LOTUS); (2) RDKit — rdkit.org; (3) Bradley, R. A.; Terry, M. E. Biometrika 1952, 39, 324–345; (4) McInnes, L. et al. UMAP, arXiv 2018.