Student Posters Repository

For the investigation of biomedical research questions, the integration of multiple omic levels has proven to give deeper insights into underlying molecular processes or disease pathophysiology. Due to the complexity of data and the lack of methods able to deal with it, data analysis is performed level-wise and results are integrated subsequently in the majority of multiomic studies. This level-by-level approach is not only very time-consuming but also neglects inter omic crosstalk of biological mechanisms. An approach that performs multi-level integrated analysis by inferring regression-based multimodal networks was proposed previously by Ogris et al. (Scientific Reports, 2021): The Knowledge guided Multi-Omic Network inference approach, KiMONo. The method leverages prior knowledge and is able to integrate any number and any kind of omic data. Unfortunately, the inference on such big data sets comes with high computational costs. We are therefore implementing KiMONo4Py, an efficient python package of the KiMONo algorithm. 

The circumgalactic medium (CGM) of a galaxy is a reservoir of multiphase gas surrounding the galaxy. The CGM is increasingly recognized for its significant role in driving the evolution of galaxies. Traditional cosmological hydrodynamics simulations fail to resolve the CGM and struggle to reproduce the observational characteristics of the CGM due to its low density. Traditional simulations achieve spatial resolutions in the low-density galactic halo that are orders of magnitude worse than the resolution they get in the high-density disk of galaxies. This problem cannot be addressed by simply increasing the resolution everywhere in the simulation as it would be prohibitively expensive to do. In my research I design more physically realistic models and techniques to resolve CGM. I develop a hyper-refinement model for CGM in a cosmological simulation. To do so, I use highly-efficient, massively-parallel simulation code, GIZMO, that is specially designed for studying galaxy evolution. I modify the GIZMO code with the physical models I developed while maintaining its efficiency and parallelization capabilities.

Type Ia supernovae (SNe Ia) have proven their importance as cosmological standardizable candles, leading to the discovery of dark energy. These explosions are believed to be the result of a thermonuclear runaway in carbon-oxygen white dwarfs found in binary systems with a companion star. However, the true nature of their stellar progenitors still remains an active area of investigation. A crucial validation of the hydrodynamical models of these explosions is the comparison of their predicted nucleosynthetic yields and synthetic spectra and light curves against observations of optical SNe Ia transients as well as supernova remnants (SNRs). A suite of hydrodynamical models is computed using the FLASH adaptive mesh refinement code. These models are run up to the homologous expansion and subsequently post-processed in the nuclear network code Torch, and the radiation transport code SuperNu to obtain their nucleosynthetic yields and synthetic spectra, respectively. Ultimately, the synthetic spectra are classified against observed events using the supernova identification code SNID.

Studying large complicated systems with chemical accuracy requires massive scaling coupled with theoretical advances. One popular approach is QM/MM, which discretizes a system into an area of interest (where Quantum Mechanical master equations are used) and a larger bulk volume (where Molecular Mechanics are simulated). Approaches which use such a conceptual meshing have computational challenges beyond those of general finite element meshes. We plan to use QM/MM to study the solvent dynamics and response of solvent molecules to a perturbation of a solute ion, specifically Fe(II) to Fe(III).  The ion is surrounded by a QM shell of water, which is embedded in an MM representation. The coupling at the boundary is theoretically validated by the SAFIRES (scattering-adapted flexible inner region ensemble separator) scheme, which ensures that the QM molecules and MM molecules do not escape their respective regions, while maintaining statistical averages. There remain computational scaling challenges which we expect to mitigate over the course of this school.

Wind energy is seeing massive growth in recent decades, and it is now acknowledged that it will be critical for combating climate change. Many onshore wind farms will likely be located on hilly terrain or near urban centres, influencing the atmospheric boundary layer properties and wind turbine wakes. Accurately predicting the changes in wind speed, shear, direction, and turbulence intensity caused by complex terrain throughout wind farms is essential to assess their power output and for optimal wind turbine control. For this purpose, the wind farm simulator WInc3d, part of the high-order finite-difference framework Xcompact3d, is used to perform Large-eddy simulation (LES) of the flow over complex terrain for high Reynolds numbers. The computational framework offers a highly efficient parallelisation strategy with "spectral-like" accurate numerical schemes on a Cartesian mesh.

Public sector nepotism, or the practice of providing jobs and promotions to relatives in the public sector, is one of the most chronic and hard-to-identify pathologies within public administrations in the developing world. However, the lack of data on bureaucrats’ public employment outcomes and family networks have prevented the analysis of this phenomenon in modern bureaucracies. 

To contribute to filling this gap, I collect and combine detailed biographical information, employer-employee records, and the mandatory but confidential disclosure of family ties for every worker in the Colombian public administration from 2011 to 2017. This extensive data collection effort, and a new methodology of family network reconstruction powered by HPC, allow me to recreate not only

1) The entire career trajectories of 1.1 million civil servants, but also

2) Their extended family networks connecting more than 2.4 million individuals via blood or marriage ties.

Using this data, I study the role of family connections to public sector managers and supervisors end up having in the allocation and compensation of public sector workers in Colombia and how this influence has been affected by the introduction of an anti-nepotism legislation reform in 2015.

We describe here the high performance computing enabling design of the electronic and optical properties of two-dimensional (2D), atomically thin, crystals. This approach uses a combination of ab initio computationally intensive density functional theory based tools, development of tight-binding (TB) approximation and the solution of Bethe-Salpeter equation (BSE) allowing for the description of many-particle optical complexes. In the first step the electronic structure from first principles is determined. Next, the analysis of Kohn-Sham wavefunctions is performed, the task that is computationally challenging due to the significant size of matrices and high numbers of integrals that need to be evaluated numerically. The process of the ab initio based TB model construction toward studies of, i.e., twisted van der Waals heterostructures and multimillion atom nanostructures, requires solving the multidimensional minimalization problem. Furthermore, the focus of this research is to develop the theoretical understanding of the optical properties of correlated electrons in 2D crystals, combining the problem of numerically challenging analysis of interactions that enter differential equations in matrix forms and further operating on large-size dense matrices that need to be diagonalized efficiently. The entire numerical procedure is based on the Fortran Intel OneApi and OpenMP routines.

Less than one percent of stars in a galaxy form with masses exceeding ten solar masses. Despite their rarity, these massive stars are believed to play a crucial role in shaping their surroundings and producing astronomical transients, ultimately determining the evolution of the star cluster or galaxy they are located in. In light of recent observations, such as the detection of gravitational waves, it has become important that the contribution of massive stars is correctly accounted for in simulations of stellar multiples and star clusters. However, for large systems such as globular clusters, which can contain millions of stars, simulating stellar interactions for populations of stars is computationally expensive and the evolution of massive stars remains highly approximated in most star cluster codes. Therefore, in my research, I am developing an improved and up-to-date treatment of stellar evolution that can be easily integrated with the star cluster codes while keeping computational overhead to a minimum. By combining the latest models of massive stars with codes that can simulate the evolution of stellar populations, we can better understand the dynamics and composition of stellar systems as well as the evolution of gravitational wave progenitors.

Simulation of air pollutant concentrations in the atmosphere is a vital task to evaluate the policies, standards, and strategies to control the emission of pollutants and global warming. Many health concerns due to poor air quality are addressed using these simulations. Chemical Transport Models( CTMs) are tools to model chemical fate and transport in the atmosphere, by numerically solving physics and chemistry equations. For a Hemispheric domain with a 12x 12 km resolution, on average 32 to 64 CPUs with 4 to 6 Gb memory per each are requested for 3 to 4 hours of run time for concentration simulation of one day. Modeling perturbations of pollutants with each of these tools, in different sources and/or receptors, is a time-consuming and computationally expensive process(even with HPC tools!). The development of an appropriate deep learning algorithm is the goal of my research to overcome this issue. For the purpose of efficiency, the developed tool will be written in a parallel mode.