Supercomputing Frontiers and Innovations

Effective Algorithms of the RI Approximation for the CIS Method: an Example of Application of the High-Memory Strategy in the Ab Initio Calculations

Maksim D. Zhukov — 2026-04-27

Two variants of the CIS methods with the RI approximation have been implemented. Both methods employ the high-memory strategy: the first, RI-CIS(1), is based on the full storage of the decomposed electronic repulsion integrals (ERI) tensor and CIS Hamiltonian, while the second, RI-CIS(2), stores only the decomposed ERI tensor. Both variants of the RI-CIS were tested for parallelism, performance, and precision. The results are compared with the default CIS method and RIJCOSX approximation. The considered methods demonstrate higher performance compared to their analogs and higher precision compared to the RIJCOSX approximation. Even the worse scaling of the RI methods did not lead to the lower performance in the conducted test calculations. The reported algorithms show that the performance of the quantum chemistry calculations is limited not only by the CPU power but also by the availability of RAM. The large volume of available memory can significantly increase the speed of the calculation by employing more effective but memory-consuming algorithms.

High-Performance Computing in the Molecular Dynamics of Tubulin Cytoskeleton Polymers

Ilya B. Kovalenko — 2026-04-27

High-performance computing is one of the most essential tools fueling the advancement of computational biology. The article discusses the application of the full-atom molecular dynamics (MD) method to study the dynamic behavior of filaments formed by the protein tubulin, and presents the results of testing the calculation performance depending on the latest models of central processors and video accelerators. Our comparative performance analysis of GPU-based computing architectures for all-atom MD simulations of biomolecular systems not only provides guidance on choosing the best computing solution in terms of price-performance ratio, but also shows the maximum potential computational performance that modern CPUs and GPUs can provide. For example, MD of the biomolecular system containing a tubulin protofilament in an explicitly specified solvent consisting of more than 300 thousand atoms can be studied with performance of 232 ns/day at time step 2 fs when using single-node computer with the latest CPU and GPU generation architecture. Constantly evolving computing resources coupled with modern software enable us to solve increasingly complex problems in life sciences.

Ionic and Water-Saturated Clusters in Self-Healing Polydimethylsiloxanes Modelled by Molecular Dynamics

Tatiana M. Makarova — 2026-04-27

Elaboration of new self-healing polymer materials with improved properties such as room-temperature healing and sufficient mechanical performance is a complicated task. In our study, we present the groundwork for construction of multicomponent systems for polydimethylsiloxane-based polymers via condensation in molecular dynamics resembling the simulated annealing protocol. Upon the self-organization according to the force field that model the intermolecular interactions, all the compounds of the “siloxane equilibration” system reproducibly assembles in a polydisperse structure with ionic and water-saturated aggregates. Around these aggregates, the negatively charged polymer terminal groups are oriented, along with ions of initiators and residual water. At the temperature of self-healing, the outer layers of the aggregates intensively exchange with each other. Thus, the molecular dynamic simulations shed light on crucial structural and dynamical properties on a molecular level that can influence the self-healing process, which would be useful in further targeted development of these materials having a prospect in cable products.

Comparative Molecular Dynamics Study of E- and Z-Biliverdin-IXα Binding to Human Serum Albumin

Igor V. Polyakov — 2026-04-27

Human serum albumin (HSA) is the main transporter of a wide range of endogenous ligands, including linear tetrapyrrolic bile pigments. Despite many experimental and theoretical studies, the detailed binding modes of linear tetrapyrroles in HSA remain not fully understood. Here, we investigate the interaction of 4Z,15E and 4Z,15Z biliverdin-IXα with HSA by classical molecular dynamics and machine learning. Starting from the crystallographic complex of HSA with 4Z,15E bilirubin-IXα, we construct models for both biliverdin isomers and explore their conformational space in the initial binding site. We analyse protein-ligand contacts, conformational flexibility, and the populations of distinct binding poses using clustering of interaction fingerprints. The results reveal both shared and isomer-specific interaction patterns between biliverdin and HSA. Several conserved contacts are maintained in both complexes, while distinct differences in contact occupancies and binding pocket conformations are observed between the E- and Z-isomers. Overall, this study provides a consistent molecular level picture of how biliverdin isomers interact with HSA and demonstrates a practical workflow for analysing flexible protein-ligand complexes by combining molecular dynamics with interaction fingerprint clustering.

MPI+OpenMP Implementation of Resolution-of-the-Identity Hartree–Fock Method Exploiting Permutational Symmetry of Three-Center Electron Repulsion Integrals

Iurii V. Kashpurovich — 2026-04-27

We report a high-performance implementation of the resolution-of-the-identity Hartree–Fock method that fully exploits the permutational symmetry of three-center electron repulsion integrals (ERIs). The present implementation adopts a hybrid MPI+OpenMP parallelization strategy. Two different algorithmic approaches (with and without the preliminary transformation of ERIs) are analyzed and compared. A custom data layout introduced previously is employed. Designed to efficiently leverage the permutational symmetry of ERIs, it minimizes not only inter-node communication but also local memory traffic. Other extensive low-level and algorithmic optimizations are proposed and discussed. Reasonable parallel scaling is demonstrated by performance benchmarks on a chlorophyll dimer (C₅₅H₇₂O₅N₄Mg)₂in an aqueous environment of 48 molecules (322 atoms overall, 3700 and 11896 functions in main and auxiliary basis sets, respectively). Peak speedups of 84× and 71× on 128 threads are achieved for the ERI calculation and the exchange matrix construction, respectively, within the algorithm involving the preliminary transformation.

pH-Dependent Conformational Analysis of Threonine Using Different Molecular Modeling Methods

Mikhail E. Kuznetsov — 2026-04-27

Conformational landscape of flexible molecules plays an important role in their reactivity, physicochemical properties and biological functions. The article presents a comparative study of the conformational stability of three protonated forms of threonine (Thr⁽⁺⁾, Thr⁽⁰⁾, Thr⁽⁻⁾) in aqueous solution using classical molecular dynamics (MD), umbrella sampling (US) and metadynamics (MTD) methods. It is shown that classical molecular dynamics fails to achieve ergodic sampling for Thr⁽⁰⁾and Thr⁽⁻⁾due to high rotational energy barriers around the C_α–C_β bond. The US method, despite being slightly more computationally expensive than classical MD, provides the most accurate Gibbs free energy profiles with minimal statistical error. Conventional MTD exhibits an unacceptably high confidence interval (up to 6 kcal/mol), while well-tempered MTD (WT-MTD) yields results that are quantitatively consistent with US (difference less then 0.2 kcal/mol) and an acceptable error margin (∼1 kcal/mol). It was established that at pH < 9.62 (Thr⁽⁰⁾and Thr⁽⁺⁾forms), the trans conformation is the most stable, whereas for the deprotonated Thr⁽⁻⁾form, the gauche⁽⁻⁾conformation is preferred. At the same time, the energy differences between the conformers are small (1–2 kcal/mol), and the transition barriers vary within the range of 3–12 kcal/mol.

High-Throughput Computational Discovery of Anti-Coronavirus Agents in the COVID-19 Era: Crucial Insights for Combating Emerging Biogenic Threats

Dmitry S. Druzhilovskiy — 2026-04-27

In May 2020, the Joint European Disruptive Initiative (JEDI) launched the “Billion Molecules against COVID 19” challenge – an extensive open science effort aimed at identifying small molecule inhibitors of SARS-CoV-2 and related human receptors. Our research group joined this initiative among 130 international teams, focusing on the in silico screening for potential anti coronavirus agents that target three viral proteins and one human receptor. The screening campaign covered more than one billion synthetically accessible structures, including approved pharmaceuticals. By July 17, 2020, our team submitted a subset of 10000 prioritized compounds to the organizers for expert evaluation. The results from our selection, together with those from 19 other participating teams, contributed to a pool of approximately 1000 molecules selected for chemical synthesis and bioactivity testing. In total, 878 compounds were successfully synthesized and evaluated for inhibitory activity against various SARS-CoV-2 targets as well as the human serine protease TMPRSS2. Ultimately, 27 compounds – including one proposed by our group – demonstrated measurable anti coronavirus activity. The collective outcomes of these collaborative efforts were reported in the “Molecular Informatics” journal in 2024. In the present study, we summarize our participation in the JEDI challenge and discuss broader methodological and organizational considerations critical for improving the efficiency of rapid scientific responses to future emerging biological threats.