Pavlo O. Kasyanov

Protein behavior inside cells is dominated by the crowded nature of the intracellular environment. Progress in structure determination of proteins and protein complexes, based on advances in Artificial Intelligence, provides an opportunity for structure-based modeling of cellular phenomena. Such modeling at the atomic resolution has been advanced by the traditional simulation techniques, e.g. molecular dynamics. A recently developed docking-based approach implements Markov Chain Monte Carlo sampling of intermolecular energy landscapes, offering several orders of magnitude faster simulation protocols. The approach allows addressing much longer trajectories of macromolecular systems in the crowded intracellular environment at atomic resolution. The sampling by design avoids low-probability (high-energy) states, which greatly accelerates the simulation process. A notable feature of this docking-based approach is the rigid body approximation of protein structures. The rigid-body approximation had been the primary direction in the protein docking field up until recent developments in deep learning. The rigid-body approach should be quite robust for the higher energy transient interactions that dominate the highly crowded cellular environment, as they likely involve relatively small conformational change. However, it is less applicable to the low-energy protein-protein complexes, especially those involving flexible regions. We addressed this problem by incorporating AlphaFold3 top models of the protein complexes in the mapping of the intermolecular energy landscape, as representative of the low-energy configurations of the protein assembly. By the nature of the AlphaFold predictions, these models involve appropriate conformational change between unbound and bound structures. These low-energy docking poses are combined with the rigid-body docking predictions that cover the multiplicity of the transient interactions. Such combination directly addresses the conformational flexibility of proteins upon binding along with the multiplicity of the transient protein encounters in the crowded cellular environment.

У доповіді наведено результати фундаментальних досліджень за новим науковим напрямом у теорії стохастичних процесів ухвалення рішень, який за координації Відділення інформатики НАН України сформовано на базі Навчально-наукового комплексу «Інститут прикладного системного аналізу» Національного технічного університету України «Київський політехнічний інститут імені Ігоря Сікорського» МОН України та НАН України. Розглянуто також кілька стратегічних та оперативних прикладних проєктів, які виконуються на основі отриманих математичних результатів і сучасних методів штучного інтелекту у співпраці з провідними українськими та міжнародними партнерами. Як цитувати: Касьянов П.О. Інтегровані дослідження в галузі штучного інтелекту, математичного моделювання та оборонних застосувань (стенограма доповіді на засіданні Президії НАН України 22 жовтня 2025 р.). Вісник НАН України. 2025. № 12. С. 56—62. https://doi.org/10.15407/visn2025.12.056

This note presents an analytical framework for decision-making in drone swarm systems operating under uncertainty, based on the integration of Partially Observable Markov Decision Processes (POMDP) with Deep Deterministic Policy Gradient (DDPG) reinforcement learning. The proposed approach enables adaptive control and cooperative behavior of unmanned aerial vehicles (UAVs) within a cognitive AI platform, where each agent learns optimal energy management and navigation policies from dynamic environmental states. We extend the standard DDPG architecture with a belief-state representation derived from Bayesian filtering, allowing for robust decision-making in partially observable environments. In this paper, for the Gaussian case, we numerically compare the performance of policies derived from DDPG to optimal policies for discretized versions of the original continuous problem. Simulation results demonstrate that the POMDP-DDPG-based swarm control model significantly improves mission success rates and energy efficiency compared to baseline methods. The developed framework supports distributed learning and decision coordination across multiple agents, providing a foundation for scalable cognitive swarm autonomy. The outcomes of this research contribute to the advancement of energy-aware control algorithms for intelligent multi-agent systems and can be applied in security, environmental monitoring, and infrastructure inspection scenarios.

This paper presents a concept for a cognitive AI platform that enables autonomous navigation of distributed multi-agent systems, exemplified by UAV swarms. The proposed architecture integrates a ground control center with cognitive services and a multi-layered onboard subsystem, supporting a continuous loop of learning, adaptation, execution, and behavioral model updates. Several core mission scenarios are introduced, such as reconnaissance, search and rescue, target neutralization, and deception, showcasing the swarm’s ability to operate autonomously and in a decentralized manner, even under adversarial conditions. An example of a search and rescue mission implementation plan using a cognitive platform that includes adaptive planning, SLAM navigation, swarm coordination, and deep object recognition is presented. The results were partially supported by the National Research Foundation of Ukraine, grant No. 2025.06/0022 “AI platform with cognitive services for coordinated autonomous navigation of distributed systems consisting of a large number of objects”.

This note presents an analytical framework for decision-making in drone swarm systems operating under uncertainty, based on the integration of Partially Observable Markov Decision Processes (POMDP) with Deep Deterministic Policy Gradient (DDPG) reinforcement learning. The proposed approach enables adaptive control and cooperative behavior of unmanned aerial vehicles (UAVs) within a cognitive AI platform, where each agent learns optimal energy management and navigation policies from dynamic environmental states. We extend the standard DDPG architecture with a belief-state representation derived from Bayesian filtering, allowing for robust decision-making in partially observable environments. In this paper, for the Gaussian case, we numerically compare the performance of policies derived from DDPG to optimal policies for discretized versions of the original continuous problem. Simulation results demonstrate that the POMDP-DDPG-based swarm control model significantly improves mission success rates and energy efficiency compared to baseline methods. The developed framework supports distributed learning and decision coordination across multiple agents, providing a foundation for scalable cognitive swarm autonomy. The outcomes of this research contribute to the advancement of energy-aware control algorithms for intelligent multi-agent systems and can be applied in security, environmental monitoring, and infrastructure inspection scenarios. The results were partially supported by the National Research Foundation of Ukraine, grant No. 2025.06/0022 "AI platform with cognitive services for coordinated autonomous navigation of distributed systems consisting of a large number of objects".

Discrete time control systems whose dynamics and observations are described by stochastic equations are common in engineering, operations research, health care, and economics. For example, stochastic filtering problems are usually defined via stochastic equations. These problems can be reduced to Markov decision processes (MDPs) whose states are posterior state distributions, and transition probabilities for such MDPs are sometimes called filters. This paper investigates sufficient conditions on transition and observation functions for the original problems to guarantee weak continuity of the filter. Under mild conditions on cost functions, weak continuity implies the existence of optimal policies minimizing the expected total costs, the validity of optimality equations, and convergence of value iterations to optimal values. This paper uses recent results on weak continuity of filters for partially observable MDPs defined by transition and observation probabilities. It develops a criterion of weak continuity of transition probabilities and a sufficient condition for continuity in total variation of transition probabilities. The results are illustrated with applications to filtering problems.

Computational approaches can provide details of molecular mechanisms in a crowded environment inside cells. Protein docking predicts stable configurations of molecular complexes, which correspond to deep energy minima. Systematic docking approaches, such as those based on fast Fourier transform (FFT), also map the entire intermolecular energy landscape by determining the position and depth of the full spectrum of the energy minima. Such mapping allows speeding up simulations by precalculating the intermolecular energy values. Our earlier study combined FFT docking with the Monte Carlo protocol, enabling simulation of cell-size, crowded protein systems with seconds, and longer trajectories at atomic resolution, several orders of magnitude longer than those achievable by alternative approaches. In this study, we present a further drastic extension of the modeling capabilities by parallelized implementation of the simulation protocol. The procedure was applied to a panel of Death Fold Domains that form nucleated polymers in human innate immune signaling, recapitulating their homooligomerization tendencies and providing insights into the molecular mechanisms of polymer nucleation. The parallelized protocol allows extension of the simulation trajectories by orders of magnitude beyond the previously reported implementation, reaching into the uncharted territory of atomic resolution simulation of cell-sized systems.