Large Language Models and Machine Learning in Biomedical and Material Sciences

Dear Colleagues,

Generative models and large language models have reached the point where they can propose candidate molecules and materials with targeted properties. The question is no longer whether these tools can generate chemically plausible structures, but whether their outputs survive contact with synthesis and measurement. This Special Issue collects research that uses LLM and ML methods to design functional molecules and materials and tests whether the designs work.

The scope is deliberately broad, including all branches of chemistry.

Topics of interest:

(1) Design of functional small molecules and materials

Property-conditioned generation; multi-objective optimization; lead optimization; catalyst and ligand design; polymer and formulation design; electrolyte and solvent design; porous and framework materials; membranes and separation materials; photochemical and electronic materials.

(2) LLM/ML-enabled synthesis and feasibility planning

Retrosynthesis planning, reaction condition design, route feasibility scoring, integration of synthesis constraints into generative design, and workflows that connect design outputs to actionable synthetic plans.

(3) Functional biomolecule design

Peptide and protein design for stability, binding, catalysis, assembly, phase behavior, and function; sequence-to-structure-to-function pipelines; constrained design under reduced alphabets; design under explicit biophysical or evolutionary constraints.

(4) End-to-end supervised discovery workflows

Pipelines where LLM/ML tools contribute across multiple stages, including problem formulation, evidence mapping, hypothesis formation, candidate proposal, computational evaluation, experimental planning, analysis, and manuscript preparation, with human verification at claim-critical points.

(5) Reliability, benchmarking, and failure-mode analysis

Benchmarks comparing design pipelines; quantification of uncertainty and robustness; documented failure modes such as chemically invalid structures, non-viable synthetic routes, spurious property predictors, data leakage, and citation errors, together with mitigation protocols.

Prof. Dr. Boggavarapu Kiran
Dr. Shun Dong
Topic Editors