99Science: Science in the age of AI — figure sources
Research and rendering date: 10 October 2026.

Requirements: Python 3, numpy, matplotlib, Pillow; PyMOL for the ribbon rendering.
From this folder: python build_figures.py; pymol -cq render_ubiquitin.pml;
then python build_beautiful.py. The second Python script creates the final six figures.

PDB coordinates: 1UBQ, chain A, https://doi.org/10.2210/pdb1UBQ/pdb.
CASP14 summary statistics: Jumper et al. (2021), doi:10.1038/s41586-021-03819-2.
Lorenz equations: dx/dt=10(y-x), dy/dt=x(28-z)-y, dz/dt=xy-(8/3)z.
Integrator: RK4, dt=0.002, 40 time units; initial conditions (1,1,1) and
(1.000001,1,1). Full time/separation data are in data/lorenz-divergence.csv.
The first trajectory's x-z projection also supplies the article's cover.

The Rosenbrock surface is evaluated directly from the displayed formula.
The odd-number tiling is an exact construction, not sampled data.
The active-learning and screening figures are explicit mathematical worked
examples, not measured AI performance or laboratory data. The Gaussian-process
kernel, observations and selection rule are recorded in active-learning-method.json.
The molecular cartoon is a rendering of experimental coordinates, not an
AlphaFold prediction; the CASP14 statistics below it are a separate benchmark.

All artwork is newly rendered for this article. Sources are credited in its captions.
