LFER explorer
The correlation from the schema document, made live. Toggle molecules in and out of the training set and the fit recomputes: coefficients, R², RMSE, significance, and the survive-controls test against molecular weight. Built to make one thing visible — how little a seven-point model can be trusted, and how the schema's guards catch it.
Illustrative values. Molecule identities and functional-group
decompositions are real chemistry. The HOMO energies and ozone rate constants are plausible
magnitudes, not curated measurements or converged runs. This explorer demonstrates
the method; it does not report results.
Try the presets in order — that is the point of this page.
All seven molecules give a fit that survives the size control (p = 0.005). Remove just
two — PFOA and sulfamethoxazole, the ends of the range — and the same relationship no
longer survives (p = 0.055), while colinearity between HOMO and weight jumps from VIF 1.08 to
5.13. Narrow the range further and it fails outright (p = 0.23). Nothing about the chemistry
changed; only which seven points were on the page. This is why the state-of-the-field note
insists the property curation, not the descriptor computation, is the real work.
log₁₀ k(O₃) vs. highest group HOMO
7 of 7 molecules in the fit · hover a point for its governing group · click to toggle
Dashed drops are residuals. Hollow points are excluded from the fit but still drawn, so the effect of dropping them stays visible.
Fitted model
Simple LFER, response log₁₀-transformed
—
R²—
RMSE (log units)—
slope ± SE—
t (slope)—
p (two-tailed)—
n—
Training set
Untick to exclude from the fit
| Molecule | HOMO | log k | resid |
|---|