A structure-aware kernel
Binary Morgan fingerprints and five standardized eluent descriptors describe the molecule and its measurement conditions.
FREE WEB UI · NO LOGIN · BROWSER CPU
From structure
to signal.
Explore electrospray ionization efficiency.
Bring a molecule. Set your conditions. Understand the uncertainty.
450benchmark compounds
1,378experimental measurements
5measurement sets
90%model-based predictive interval
01 / PREDICTION WORKSPACE
A molecule, a method, a prediction.
Initializing model…
SMILES notation
Try a molecule
No file selected
Download a CSV template ↗Each row uses its own eluent descriptors. The selected measurement context applies to the whole batch.
Descriptor values use the original benchmark scales. Use a reference preset or enter values calculated in those same scales.
A new method includes uncertainty in the measurement-set offset. Add measured standards to adapt the prediction to your method.
Add calibration standards OPTIONAL +
Upload standards measured with your new method. Include SMILES, all five eluent descriptors, and measured logIE. These observations condition the GP posterior; they do not refit its hyperparameters.
No calibration standards added
◈ Your molecules and standards stay in this browser. Prediction inputs are not stored.
PREDICTION OUTPUTHGP · v1.0
ie−
The next signal
starts with a structure.
Enter your molecule and eluent conditions.
Your predicted efficiency, uncertainty, and
structural similarity will appear here.
01 STRUCTURE02 CONDITIONS03 SIGNAL
PREDICTED log₁₀ IE
—
90% predictive interval
Includes observation noise. Model-based; coverage is not conformal-calibrated.
Structural similarity
Maximum Tanimoto similarity to training compounds · Morgan radius 2 / 2,048 bits
BATCH COMPLETE0 molecules.
0 molecules.
Every signal accounted for.
| Molecule | logIE | 90% interval | Similarity |
|---|
02 / BEHIND THE PREDICTION
Model the chemistry.
Account for the method.
A hierarchical Gaussian process combines molecular similarity and eluent conditions with a measurement-set effect. Calibration standards help adapt predictions to a new analytical method.
Calibration with context
Measured standards update the posterior for your new method. A set-level offset and chemical similarity both affect the update.
Uncertainty you can inspect
A model-based interval and structural similarity provide context. They do not guarantee accuracy for unfamiliar chemistry or remove missing speciation information.
Scientific scope, provenance & reproducibility +
Full-benchmark model. Hyperparameters come from the project's full-data fit on 1,378 measurements across 450 compounds. The deployed model is distinct from the paper's leave-one-set-out test models; their test metrics cannot be assigned to this web tool. Queries that match training structures are explicitly flagged.
Calibration and intervals. User standards condition the joint Gaussian posterior with fixed hyperparameters. Intervals are 90% Gaussian predictive intervals, including observation noise and uncertainty in a new set effect. No conformal calibration or pKa/speciation correction is applied.
Data source. Benchmark released with Wang et al., Analytical Chemistry 2025, 97, 13131, at kruvelab/MS2Quant ↗. Original raw files are retrieved directly by your browser from the authors' public source and hash-verified at initialization. The benchmark posterior is cached locally for faster return visits; your prediction inputs are not cached. This Space does not redistribute the raw data.
Use the correct scale. Negative-mode ESI only. logIE is relative ionization efficiency on the benchmark scale, not a concentration or an absolute response factor. A benchmark set option refers to that exact measurement context, not every molecule of the named chemical class.
Companion manuscript. What Can Calibration Standards Fix? Measurement-Domain and Chemical-Domain Shifts in Transferring Electrospray Ionization Efficiency Models. No publication status is implied.