Attic Molecular Discovery Lab (2D)

Four stations as flat maps: drag a ligand over a pocket energy map, fit an IC50 through noisy dilutions, heat and pull a molecule, and count cells with Poisson error.

Switch to 3D

Presets

Playback

Actions

Parameters

Layers

Model check

The science

Docking pocket (station 1)

The ligand is a rigid 2D set of atoms placed at (x, y, theta) in a fixed horseshoe of protein atoms. The pseudo-binding score is a sum over ligand-pocket atom pairs: E = sum 4 eps [(sigma/r)^12 - (sigma/r)^6] + 332 qi qj /(4 r^2) x 0.4 - 1.6 exp(-((r - 2.9)/0.4)^2) (Lennard-Jones, distance-dependent dielectric Coulomb, Gaussian hydrogen-bond well for donor/acceptor pairs) plus a frozen-rotor entropy penalty. Kd = exp(dG / RT) at 1 M standard state, so every 1.36 kcal/mol is a factor of 10 in Kd at 298 K. The constants are illustrative and the scoring is not a validated predictor of real affinity; real docking is 3D, flexible and uses solvation terms.

Dose-response assay (station 2)

Each transfer of volume Vt into diluent Vd multiplies the concentration by r = Vt/(Vt+Vd); with independent volume errors, SD(ln Ck) = sqrt(k) (1 - r) CV sqrt(2). Activity follows the Hill equation y = bottom + (top - bottom)/(1 + (c/IC50)^h), fitted by Levenberg-Marquardt. The fit uses the nominal concentrations (the experimenter cannot know the actual ones), so pipetting error biases and widens the fitted IC50. Cheng-Prusoff for a competitive inhibitor: IC50 = Ki (1 + [S]/Km).

Molecular dynamics (station 3)

4-aminophenol with harmonic bonds, angles, out-of-plane terms and Lennard-Jones non-bonded pairs, integrated with velocity Verlet (dt = 0.5 fs) and an exact Ornstein-Uhlenbeck Langevin thermostat. Equipartition says <1/2 m vx2> = 1/2 kB T, so velocities scaled by sqrt(kB T/m) follow a standard normal distribution (bottom-right graph). The force field is a toy, tuned for look and stability, not a production force field. The view is the x-y projection of the 3D molecule.

Hemocytometer (station 4)

Each large square holds 0.1 uL (1 mm x 1 mm x 0.1 mm), so the expected count is lambda = C V / dilution and counts are Poisson distributed: mean = variance = lambda, relative error of a total count N is 1/sqrt(N). Counting about 400 cells gives roughly 5 %. A systematic miscount (clumps, missed cells) biases the estimate low and no amount of counting removes it.

All numbers come from the same sim_core module as the 3D page. The independent checks above run in your browser: Lennard-Jones minimum, dG-Kd round trip, Cheng-Prusoff plus fit recovery, Poisson mean/variance, and the C-O oscillator period against 2 pi sqrt(mu/2k). Fictional lab; educational only, not for real drug design or clinical decisions.

a
b
c
d