TYC Soiree: Modelling and simulation of biomolecules: pathways, kinetics and catalysis

15:00 – 15:40	Gabriele Corso, Massachusetts Institute of Technology
15:40 – 16:20	Marc Van der Kamp, University of Bristol
16:20 – 17:00	Stefano Motta, University of Milano Bicocca
17:00 onwards	Drinks reception at UCL Physics E7

Boltz: Towards Accurate Biomolecular Modeling and Design – Gabriele Corso, Massachusetts Institute of Technology

Accurately modeling biomolecular interactions remains a central challenge in modern biology. Breakthroughs such as AlphaFold3 and Boltz-1 have greatly advanced structure prediction of biomolecular complexes. Building on this progress, we introduced Boltz-2, the first AI model to approach the accuracy of free-energy perturbation (FEP) methods for estimating small molecule–protein binding affinities. Most recently, with BoltzGen, we demonstrated that fine-tuning large-scale structure prediction models for protein design enables a powerful end-to-end pipeline. We validated this pipeline experimentally with multiple wet-lab collaborators, achieving successful designs across a wide range of novel targets.

Gabriele Corso recently received his PhD from MIT CSAIL where his research focused on developing novel ML frameworks to tackle challenging problems in drug discovery and he led the development of popular models in the space including DiffDock, Boltz-1 and Boltz-2.

Mapping Biomolecular Conformational Pathways with Self-Organizing Maps – Stefano Motta, University of Milano Bicocca

Understanding complex biomolecular processes from molecular dynamics (MD) simulations requires interpreting large, high-dimensional datasets. I will discuss how Self-Organizing Maps (SOMs), a type of unsupervised machine learning models, can generate intuitive, low-dimensional representations of the conformational space sampled during these simulations (1). I will then demonstrate how this approach, implemented in our new R package SOMMD(2), can be used to reconstruct molecular pathways in processes such as protein unfolding and ligand binding, and to build transition network models that characterize their key events(1,3).

1.      Motta, S., Callea, L., Bonati, L., & Pandini, A. (2022). PathDetect-SOM: A Neural Network Approach for the Identification of Pathways in Ligand Binding Simulations. Journal of Chemical Theory and Computation, 18(3), 1957–1968.

2.      Motta S., Callea L., Mulla S. I., Davoudkhani H., Bonati L., Pandini A. (2025). SOMMD: an R package for the analysis of molecular dynamics simulations using self-organizing map. Bioinformatics, 41(6), btaf308.

3.      Callea L., Caprai C., Bonati L., Giorgino T., Motta S. (2024). Self-organizing maps of unbiased ligand–target binding pathways and kinetics. The Journal of Chemical Physics, 161, 135102.

Stefano Motta obtained his PhD in Chemical Sciences from the University of Milano-Bicocca in 2018, where he has been an Assistant Professor since 2022. His research focuses on the development and application of computational methods to investigate the structure and dynamics of biomolecular systems. His current research interests include the use of molecular dynamics to study protein-ligand recognition, the mechanism of action of bHLH-PAS proteins, the characterization of nanosystems for biomedical applications, and the development of machine learning approaches for the analysis of complex simulations.

Multiscale simulations for understanding and engineering enzymes: from QM/MM to ML/MM – Marc Van der Kamp, University of Bristol

Enzymes have excellent potential as selective and efficient biocatalysts for industry. Obtaining enzyme biocatalysts with both the desired selectivity and activity, however, remains a challenge. Atomistic simulations can provide valuable information for rational engineering. Ideally, simulation protocols require limited computational resource (and thus energy) but maintain sufficient accuracy. Here, the development of tools and methods for ‘in silico enzyme screening’ with reaction simulations are discussed, highlighting applications to different enzymes. We have shown that short QM/MM reaction simulations with semi-empirical QM methods can be used to correctly indicate activity for certain enzymes, such as serine beta-lactamases.¹ When combined with automated protocols to set up simulations, such as Enlighten2,² this can result in efficient evaluation of activity and selectivity. We demonstrate how this can be used to obtain key insights into natural beta-barrel Diels-Alderases,³ which are promising stable and stereoselective biocatalysts. 

For highly efficient screening of enzyme activity, such that calculations can be used during enzyme (re)design, further increases in efficiency and accuracy are important. Replacing QM by machine-learning (ML) potentials can, in principle, offer QM accuracy at a fraction of the computational cost. However, due to the absence of electrons in ML potentials, properly describing the electrostatic interaction between ML and MM regions, crucial for capturing enzyme catalysis, is a challenge. We have developed the “electrostatic ML embedding” (EMLE) scheme that solves this issue, allowing DFT/MM accuracy.^4,5 Here, we show that this method can be applied for enzyme reaction simulations to capture key catalytic effects. 

References: 

1. V. H. A. Hirvonen, K. Hammond, E. I. Chudyk, M. A. L. Limb, J. Spencer, A. J. Mulholland and M. W. van der Kamp. J. Chem. Inf. Model. 2019, 59, 3365-3369.
2. K. Zinovjev and M. W. van der Kamp. Bioinformatics, 2020, 36, 5104–5106.
3. L. Maschio, et al. Chem. Sci. 2024, 15, 11572-11583; Mbatha et al., Chem. Sci. 2024, 15, 14009-14015.
4. K. Zinovjev, L. Hedges, R. M. Andreu, C. Woods, I. Tuñón and M. W. van der Kamp. J. Chem. Theory Comput. 2024, 20, 4514–4522.
5. V. Gradisteanu, E. W. Chan, L. Hedges, M. Malagarriga, R. David, M. de la Puente, D. Laage, I. Tuñón, M. W. van der Kamp, K. Zinovjev. ChemRxiv 2025, DOI: 10.26434/chemrxiv-2025-nw9lt. 

Marc is Associate Prof. in Computational Biochemistry in the School of Biochemistry in Bristol, and is an expert in biomolecular simulation of enzymes and their reactions. After obtaining a PhD in this field in Bristol (2008), he pursued postdoctoral research at the University of Washington (with Prof. Valerie Daggett) and in Bristol (with Prof. Adrian Mullholland). Then, as a BBSRC David Phillips Fellow (2015-2021), he established a group with PDRAs and PhDs and advanced the use of detailed biomolecular simulation for understanding enzyme biocatalysts and predicting properties of their variants. The main research interests in the group include: enzymes involved in antibiotic resistance, computational simulation methods to aid enzyme engineering and design of biologic drugs, and further understanding the principles of enzyme catalysis and specificity.