Explore the secret of life to shape the future

Research

Our aim is to understand, at a high level of detail, how the interactions of proteins are encoded in their sequences and structures, exploring the specificity for protein engineering and protein-based therapeutics, in research programs that combine bioinformatic analysis, structural modeling, computational design and experimental characterization. To address this, our research centers on the following themes: protein-based therapeutics, protein function, protein design, physics-informed intelligence, functional genomics, and applied mathematics in biology.

Protein-based therapeutics

How to develop the next generation therapeutics?

Discover transformative molecules through cutting-edge artificial general intelligence and translate them into therapies for patients.

Protein-based therapeutics: machine learning has emerged as a powerful tool for advancing the development of protein therapeutics in medicine, leading to a revolution in the field of drug discovery. Specifically, we are developing novel algorithms to design proteins with (1) enzymatic or regulatory activity, (2) specific targeting activity, and (3) vaccines. The ultimate goal is to engineer these proteins to treat various human diseases, thereby offering new treatments for patients in need.

Missense effects: quantifying the pathogenicity of genetic variants in human disease-related genes has the potential to advance evidence-based clinical decision making and transform healthcare. We are developing physics-/evolution-informed machine learning approaches to address the knowledge gap.


Physics-informed intelligence

How to develop cutting-edge technologies in artificial intelligence?

Physics-informed learning: we develop explainable artificial intelligence using information obtained by the physical laws for prediction of disease variants and pinpoints of protein function.

Probabilistic deep learning: we build deep learning-based models with probabilistic modules for representing and processing uncertainty in both data and models.


Relevant publications:

  • Jang, Y. J., Qin, Q.-Q., Huang, S.-Y., Peter, A. T. J., Ding, X.-M., and Kornmann, B.. High-accuracy protein function annotation using physics-informed graph networks. (2023)
  • Xu, Y. C., ShangGuan, T. J., Ding, X. M., and Jang, Y. J.. Accurate prediction of protein torsion angles using evolutionary signatures and recurrent neural network. Scientific reports 11(1), 21033 (2021).

Functional genomics

What is the mathematical relationship between genes (genotype) and their effects (phenotype)?

Lipid transport proteins: intestinal lipid transport is a multistep process that requires the coordinated regulation of a series of pathways that lead to the transport of lipolytic products and micellarized lipid across the brush-border membrane, through vectorial delivery through the apical cytoplasmic compartment to the endoplasmic reticulum (ER).

Functional genomics: we identify essential genes and protein domains from SATAY data using our physics-informed models.


Relevant publications:

  • Covill-Cooke, C. et al. Shared structural features of miro binding control mitochondrial homeostasis. The EMBO Journal (2023).
  • Peter, A. T. J., Jang, Y. J., and Kornmann, B.. Csf1: a putative lipid Transport Protein required for Homeoviscous Adaptation of the Lipidome. Contact 5, 25152564221101974 (2022)
  • van Schie S., Peter, A. T. J., Jang, Y. J., Michel, A., Peter, M., and Kornmann, B.. Rewiring phospholipid biosynthesis reveals resilience to membrane perturbations and uncovers regulators of lipid homeostasis. The EMBO Journal 41, e109998 (2022).

Physics-informed learning in synthetic biology

How to advance synthetic biology using physics-informed learning?

We are employing physics-informed machine learning to create novel proteins for diagnostics & therapeutics.

Learn to design protein: amino acid sequences contain necessary evolutionary information that specifies their structures and functions. We are developing methodologies that utilize the signatures derived from this information and aim to leverage them in designing proteins with specific functions.

Quantum mechanics in biology: we seek to model biological interactions in light of quantum mechanical effects and understand aspects of biology that cannot be accurately described by the classical laws of physics.


Relevant publications:

  • Jang, Y. J.. Probabilistic learning of functional residue communities for protein design. (2023).

Molecular function

How to annotate molecular function at a high level of detail?

Protein function annotation: we develop physic-informed learning-based methods to characterize protein functional sites at the residue level. Protein function is exquisitely dependent on compactly folded structures that combine energetic stability with intrinsic flexibility. We are now trying to define the thermodynamic and evolutionary origins of metamorphic folding.

Molecular recognition: the interaction (promoting or disrupting) between two proteins often controls biological signals. We leverage computational approaches on three systems: (1) binding mechanisms of GPCR-G protein; (2) T cell receptor (TCR) activation, important in the defense against cancer; and (3) chemokines, directly implicated in human diseases.

Relevant publications:

  • Jang, Y. J.. Evolutionary signatures drive fold-shifting in metamorphic proteins. (2023).
  • Jang, Y. J., and Huang, S. Y.. Residue communities reveal evolutionary signatures of γδ T-Cell receptor. bioRxiv, 2022.12.29.522230 (2022).
  • Jang, Y. J., Peter, A. T. J., and Kornmann, B.. Leri: a web-server for identifying protein functional networks from evolutionary couplings. Comput. Struct. Biotechnol. J. 19, 3556-3563 (2021).

Applied mathematics

What can we learn from nature for developing technologies to transform healthcare?

Control theory: we develop models & algorithms that can drive the system to a desired state from given inputs by ensuring a level of control stability and achieving a degree of optimality.

Optimization: we build highly efficient optimization algorithms for the applications in all quantitative disciplines from computer science and engineering to operations research, economics, and biology.

Robotics: we design machines that can help and assist humans using the developed models and algorithms in our group.

Relevant publications:

  • Jang, Y. J., Ding, X.-M., and Shen, H.-B.. A non-homogeneous cuckoo search algorithm based-on quantum mechanism for real parameter optimization. IEEE Transactions on Cybernetics 47, 391–402 (2017).
  • Jang, Y. J., Ding, X.-M., and Shen, H.-B.. OptiFel: a convergent heterogeneous particle swarm optimization algorithm for Takagi–Sugeno fuzzy modeling. IEEE Transactions on Fuzzy Systems 22(4), 919-933 (2013).