Complex Systems¶

This example category covers larger or more chemically structured systems where interpretation depends on local context.

Representative Use Cases¶

catalyst families
heteroatom-rich molecules
scaffold-conserved series
molecules with multiple competing functional regions

Representative Surfacia Scenarios¶

The following problem types are especially well matched to Surfacia:

sulfur-containing modifier studies where one element is chemically central
homologous catalyst series where a shared fragment defines the chemistry
diverse property datasets where local and global effects compete

How to Decide Which Mode to Try¶

Use this quick rule of thumb:

if one element is already central, start with Mode 1
if one fragment or scaffold is central, start with Mode 2
if the chemistry is broad or unclear, start with Mode 3

Suggested Working Sequence¶

For a more complex project, a good strategy is:

run one broad baseline analysis
inspect the descriptor families that matter most
rerun with a more hypothesis-aware mode if the chemistry suggests it
compare compact retained feature sets rather than only comparing scores

Result Interpretation Template¶

For a complex system, this short reading pattern is often helpful:

representation Did you choose a mode that matches the chemistry?
retained descriptors Are the retained descriptors local, global, or mixed?
coherence Do those descriptors tell one story or several disconnected stories?
stability Would you still trust the interpretation if the held-out split changed?

How to Approach Them¶

For complex systems, start by deciding whether your chemistry is best represented by:

an element-specific view
a fragment-specific view
a broader exploratory LOFFI view

In larger systems, this choice often matters more than fine-tuning the model first.

Why They Matter¶

These systems are where multi-scale analysis becomes especially valuable because whole-molecule descriptors alone are often not enough.

Typical Payoff¶

Complex systems are often where Surfacia becomes most useful, because it can connect:

global size and shape
local electronic structure
fragment-level interpretation
compact model outputs that still support chemical reasoning

What to Watch For¶

Complex systems are also where misleading interpretation can happen most easily.

Be careful when:

the test set is tiny
one experimental condition dominates the model
many descriptors survive but do not form a coherent story
the representation does not match the chemistry of the dataset

What Often Works Well¶

Strong complex-system analyses often show:

a representation choice that clearly fits the chemistry
a compact feature subset rather than a diffuse large model
descriptors spanning the scales that matter for the problem
SHAP trends that can be translated into mechanistic language