Search

The even chromosome segregation between daughter cells during mitosis is crucial for genome integrity and is mostly regulated by proper attachments of spindle microtubules to kinetochores. Abnormalities in this process can lead to chromosome mis-segregation and potentially result in severe developmental disorders such as aneuploidy and cancer. Merotelic attachments when tubulin microtubules captured by the kinetochore of one chromatid originate from both spindle poles are considered as one of the key molecular processes that cause such abnormalities. In this paper, we use computer modeling and the Monte Carlo approach to reveal the reasons for retaining merotelic attachments at the end of metaphase. To this end, we varied, in small increments, the basic cell parameters within ensembles of 100, 500, and 1000 virtual cells. The analysis of configurations that ensure the preservation of the largest fraction of merotelic attachments enabled us to conclude that only a change in the size of the kinetochore corona can significantly increase the number of merotelic attachments and the angle between the centromere axis and the spindle axis. The effect of the other changes in model parameters, if any, was steadily suppressed by the end of metaphase. In addition, our computer model was validated by successfully reproducing the results of third-party theoretical studies as well as some experimental observations. We also found that the orientation of chromosomes and the number of merotelic attachments do not have an explicit correlation with each other and within some limits can change independently.

Modern architectures of central processors, in particular, AMD Zen 2 and Intel Cascade Lake, allow one to build shared memory systems with more than 100 computational cores. This paper presents the results of comparing the performance of these architectures shown on numerical modeling of mitosis in eukaryotes. The MiCoSi software that was developed by the authors and previously demonstrated a linear scalability when executed on cluster systems was used as a benchmark. The testing was performed on Amazon EC2 cloud nodes with 96 logical cores each. It is shown that, in relation to the problem of mitosis modeling, the two architectures under study have similar performance, while in some cases, Intel Cascade Lake bypasses its competitor AMD Zen 2 by 5–23%.

The paper presents the results of the implementation of computational algorithms of hydrodynamics for using graphics processor units. The implementation was carried out on the basis of the in-house CFD code SigmaFlow. Numerical simulations were based on the solution of the Navier-Stokes equations using SIMPLE-like procedure. The discretization of the differential equations was based on the control volume method on unstructured mesh. In the case of multiple CPU/GPU, parallel calculations were performed by means of domain decomposition. In the GPU-version of the code, basic computational functions were implemented as CUDA kernels to perform on GPUs. The code has been verified using several test cases. The computational efficiencies of several GPUs were compared with each other and that of modern CPUs. A modern GPU can increase the calculation performance of CFD problems by more than two times compared to a modern six-core CPU.

When the program is being optimized for execution on GPU, one has to introduce a lot of performance affected constants that define blocks parameters, data chunks size, parallelism granularity, etc. And the more software is optimized, the more magic constants it introduces. Furthermore, adding multi-GPU system support often requires usage of smart load balancing strategies that considers GPU-specific effects such as potential speed-up from ignoring some accelerators, time volatility of GPU-performance and others. As a result, performance of the target software can be significantly increased just by tuning to the hardware and processing data being used. The authors developed a means of determining the optimal values of these constants called ttgLib autotuner which is capable of monitoring the software at runtime and automatically tuning magic constants as well as performing dynamic load balancing between CPU and multiple GPUs. The performed tests showed the additional speedup upto 50-80% by tuning alone.