Basic Concepts

Understanding the core ideas behind Surfacia will help you use the framework more effectively and interpret results in a chemically meaningful way.

The Interpretability Problem

Many machine-learning models in chemistry can predict well but remain difficult to explain in a way that is useful for molecular design.

Why This Matters

In practical molecular discovery, chemists usually want more than prediction alone.

  • Which structural features drive the response?

  • Why does a modification help or hurt?

  • How can the result guide the next round of design?

Why Surfacia Focuses on Molecular Surfaces

Surfacia is built around the idea that many chemically important processes are expressed at the molecular surface.

  • Recognition depends on surface complementarity

  • Reactivity depends on local electronic structure

  • Accessibility depends on size and shape

  • Property trends often reflect a balance of global and local surface effects

Key Surface Properties

Surfacia extracts three central surface-electronic quantities.

Local Electron Attachment Energy (LEAE)

Reports local electron-accepting tendency and helps identify electrophilic character.

Electrostatic Potential (ESP)

Describes surface charge distribution and highlights electron-rich or electron-poor regions.

Average Local Ionization Energy (ALIE)

Reports local resistance to electron removal and helps identify electron-donating regions.

Multi-Scale Descriptor Design

Surfacia organizes descriptors across three hierarchical levels so that the final model can remain interpretable.

🔬 Atomic Level

Local atom-centered statistics across the molecular surface.

  • site-specific reactivity

  • atom-level extremes and averages

  • local electronic variation

🧩 Functional Group Level

Descriptors summarized over chemical fragments or automatically detected groups.

  • fragment-centered interpretation

  • modular structure-property reasoning

  • comparison between functional units

🌐 Molecular Level

Global descriptors for overall molecular character.

  • size and shape

  • whole-molecule electronics

  • bulk surface behavior

Choosing the Right Analysis Mode

Surfacia provides three complementary modes so that the descriptor strategy can follow the chemistry of your problem.

Mode 1: Element-Specific

Best when one element is already known to be mechanistically important.

  • element-aware hypothesis testing

  • heteroatom-focused studies

  • compact descriptors centered on a chosen element

Mode 2: Fragment-Specific

Best when a catalytic core, scaffold, or pharmacophore is known in advance.

  • user-defined fragment analysis

  • scaffold-conserved series

  • substituent perturbation around a fixed motif

Mode 3: Automated LOFFI

Best when no strong mechanistic prior is available.

  • automated atom-level and functional-group-level analysis

  • exploratory studies on diverse datasets

  • broad discovery without predefining a key motif

The Workflow in Plain Language

Surfacia is designed as an end-to-end pipeline from molecular structures to interpretable design insight.

        graph LR
    A[Input Structures] --> B[3D Generation]
    B --> C[QM Calculation]
    C --> D[Surface Property Mapping]
    D --> E[Descriptor Generation]
    E --> F[Compact Modeling]
    F --> G[SHAP Interpretation]
    G --> H[Interactive Analysis]

Typical stages:

  1. Build or read 3D structures from the input table

  2. Run quantum calculations and generate wavefunction outputs

  3. Extract molecular, atom-level, element-specific, or fragment-specific descriptors

  4. Select compact feature subsets

  5. Train interpretable predictive models

  6. Explain predictions with SHAP and optional language-model-assisted summaries

Compact Models, Not Just Large Feature Matrices

Surfacia is designed to reduce descriptor dimensionality while keeping the final model chemically readable.

  • Consensus-driven feature selection avoids over-reliance on one split or one ranking

  • Compact descriptor subsets are favored over oversized feature matrices

  • SHAP analysis preserves both global and sample-level interpretation

  • Trend fitting can distinguish linear, threshold-like, and saturating relationships

Descriptor Naming Guide

Feature names are intentionally systematic so users can infer their meaning directly.

  • Atom_*: atom-level global statistics over the molecular surface

  • Fun_*: functional-group-level statistics from automated grouping

  • Fragment_*: descriptors for a user-defined fragment

  • element-prefixed terms: descriptors aggregated around a selected element

  • *_min / *_max: chemically extreme regions

  • *_mean / *_average: overall character

  • *_delta: heterogeneity across sites or groups

What Makes Surfacia Different

Surfacia is meant to be more than a prediction script.

  • It keeps descriptors tied to recognizable physicochemical meaning

  • It supports hypothesis-aware and hypothesis-free analysis

  • It emphasizes compact models that remain chemically interpretable

  • It connects numerical outputs to design-oriented explanation

Next Steps

  1. Try the Quick Start page for a minimal working workflow

  2. Read Molecular Descriptors Reference for descriptor families and naming

  3. Explore Command Reference for command-level details

  4. Use Tutorials when you want guided examples