Emerging interdisciplinary directions

Materials design, inverse molecular design

How computational and machine-learning methods search backward from desired molecular or materials properties to candidate structures, with chemistry, synthesis, and uncertainty constraining the search.

IntuitionDesign backward from a useful property

Forward prediction asks what properties a structure may have. Inverse design starts with a target—such as a color, conductivity, binding affinity, or stability—and seeks structures that could meet it. Many structures may satisfy the same target, or none may satisfy all constraints.

SchoolA search space of chemical structures

Molecular graphs encode atoms as vertices and bonds as edges. Chemical validity requires more than graph syntax: valence, charge, aromaticity, stereochemistry, and stability all matter, and a valid graph may still be unsynthesizable.

UndergraduateOptimize an objective under constraints

Here $x$ denotes a candidate, $f$ a predicted objective, and $g_i$ constraints such as toxicity, cost, or stability. A model’s score is only as meaningful as its training data, calibration, and relevance to the intended experimental measurement.

AdvancedSearch, uncertainty, and validation

Bayesian optimization can choose the next candidate by balancing predicted performance with uncertainty, especially when experiments are costly. Its benefit depends on a useful surrogate model, an acquisition rule suited to the objective, and honest handling of out-of-domain candidates.

A generated structure should be checked independently for chemical validity, novelty, predicted property uncertainty, and synthetic accessibility. Retrosynthetic estimates are filters, not guarantees: route quality and laboratory success require experimental judgment.

ResearchResearch frontier

A credible discovery cycle makes the objective measurable, proposes diverse candidates, tests the most informative safe experiments, reports negative results, and updates the model and constraints. The evidence is the measured material—not the novelty of the generated structure.