From SMILES to σ-profile: the full pipeline, step by step

A SMILES string is a 1D description of connectivity. A σ-profile is the product of a 3D quantum-chemical calculation. Bridging the two correctly is where most of the engineering — and most of the opportunities to get it wrong — actually live. Here is the full pipeline, the way a production system runs it.

1. Parse and standardize the structure

The SMILES is parsed into a molecular graph, then standardized: aromaticity perceived, tautomer/charge state resolved, explicit hydrogens added. This step matters more than it looks: the σ-profile of a neutral acid and its conjugate base are completely different, so the protonation state you start from determines the physics you end up with.

2. Embed a 3D geometry

A 3D conformer is generated from the graph (distance-geometry embedding followed by a cheap force-field cleanup). For rigid molecules one conformer is enough. For flexible molecules it is not — which leads to the next step.

3. Sample conformers (when flexibility matters)

Flexible molecules populate many shapes in solution, and different shapes expose different surface charge. A single conformer can badly misrepresent the real, Boltzmann-averaged σ-profile. A proper pipeline generates a conformer ensemble, optimizes each, and Boltzmann-weights the resulting profiles by free energy. The weighting choice (temperature, energy model) is part of the recipe and should be recorded.

4. Optimize the geometry at the quantum level

The conformer is relaxed with a quantum method so the bond lengths and angles are physically correct — force-field geometries are a starting guess, not the final answer. The relaxation is run in the presence of the implicit solvent, because solvation changes the preferred geometry of polar molecules.

5. Solve the electronic structure with COSMO

This is the heart of it. A DFT calculation is run with a conductor-like implicit solvation model (COSMO), which surrounds the molecule with a virtual conductor and computes the screening charges on the molecular cavity surface. Three choices dominate the result:

• Functional — controls how electron density (and therefore polarization) is described.
• Basis set — controls how flexibly the wavefunction can respond; too small and polar groups are mischaracterized.
• Cavity construction and radii — controls where the surface sits, which directly scales every σ value.

The mfsig production recipe (v0.91.1) fixes these to a validated combination — DFT with COSMO implicit solvation on a def2-SVP basis — so that results are consistent and comparable across molecules.

6. Integrate the surface into a σ-profile

The COSMO calculation yields thousands of small surface segments, each with an area and a screening charge density σ. These are smoothed (to account for the fact that real solvent molecules are finite, not point charges) and binned into the histogram that is the σ-profile. Optional derived quantities — σ-moments, hydrogen-bond donor/acceptor splits, and the σ-potential — are computed from the same surface.

7. Stamp provenance

A σ-profile without provenance is a number you cannot defend. A reference-grade pipeline records the exact recipe, code versions, and input geometry, and signs the output so it is tamper-evident. That is what makes the result usable in a regulated or audit-bound workflow rather than just an exploratory one.

Where it can go wrong

• Wrong protonation/tautomer state — silently changes the physics.
• Single conformer for a floppy molecule — biased profile.
• Basis set too small — polar wings mischaracterized.
• Inconsistent cavity radii between molecules — profiles not comparable.
• No provenance — irreproducible, unauditable.

The value of a hosted pipeline is that these decisions are made once, validated, and held fixed — so a SMILES in always maps to a defensible σ-profile out.