Research

Intrinsically disordered proteins and regions populate dynamic conformational ensembles rather than one fixed structure. Their behavior depends on sequence composition and patterning, electrostatic state, molecular partners, and the surrounding environment. We use comparative bioinformatics and molecular analysis to study how these factors influence recognition, regulation, evolution, and disease.

This work includes large-scale disorder analysis and the development of RIDAO, while extending toward a broader understanding of functional and context-dependent disorder. A central question is how one sequence can support distinct biological roles under different conditions.

Protein fate emerges from competition among molecular association, liquid–liquid phase separation, aggregation, degradation, and cellular quality-control pathways. We study how intrinsic disorder, sequence patterning, electrostatics, concentration, and environmental conditions alter the balance among these outcomes.

A particular interest is the formation and metastability of biomolecular condensates and membraneless organelles, including how functional assemblies reorganize, mature, or transition toward less reversible states. Related work considers how proteostasis pathways respond when proteins fail to assemble or degrade properly, including emerging questions in ribosome-associated quality control.

Molecular simulation provides a direct route to studying protein motion and interactions, but many biologically important systems challenge fixed-state descriptions and conventional sampling. Prior work on machine-augmented molecular dynamics explored learned velocity updates within a physically constrained integration framework.

Current method development focuses on molecular simulations that account for changes in protonation and electrostatic state. The broader objective is to improve the treatment of flexible proteins and heterogeneous environments, where local chemistry and conformational behavior are interdependent.

Protein language models capture information about sequence and evolution, but useful downstream predictions require careful supervision, evaluation, and interpretation. We develop residue-level and protein-level models for electrostatics, ligandability, intrinsic disorder, and related functional properties.

This work includes KaML, AiPP, and the PLMPG development framework, together with methods for weak supervision, dataset refinement, and leakage-aware evaluation. The aim is to distinguish what is encoded by sequence from what changes with molecular context and to connect predictive models with physical mechanisms.