Abstract Designing drug-like molecules that satisfy multiple objectives—such as high binding affinity, synthesizability, and drug-likeness—poses a complex global optimization problem over an astronomically large chemical space. Existing deep learning-based molecular generative models often treat this task as distribution modeling, relying on atom-level autoregressive actions with less consideration of explicit optimization feedback. Consequently, they frequently generate invalid structures, converge to local optima, or produce synthetically infeasible candidates. Here, we introduce SHARP (Synthesizable Hierarchical Action-space Reinforcement learning for Pareto optimization), a molecular generator that addresses these limitations via a fragment-based hierarchical action space and reinforcement learning. SHARP ensures synthetic accessibility by applying action masks guided by a pretrained Synthesizability Estimation Model (SEM). The reinforcement learning (RL) policy is trained using a composite reward function integrating docking scores, pharmacophore matching, and solvent accessibility to generate functionally relevant and experimentally tractable molecules. Furthermore, across four lead optimization tasks—fragment growing, linker design, scaffold hopping, and sidechain decoration—on a diverse receptor set, SHARP consistently outperforms prior methods in producing molecules at high affinity and synthesizability. These results demonstrate that reinforcement learning with a chemically intuitive action space design can be an effective solution to the optimization challenges in AI-driven drug discovery, offering a robust framework for rational molecular design in structure-based applications.
Lattice cryptography is one of the promising directions in modern cryptography research. Digital signatures and key encapsulation mechanisms on lattices have already been used in practice. In the future, such quantum-resistant transformations on lattices replace all standards that are not resistant to attacks on quantum computers. This makes the analysis of their security extremely relevant. Analysis of the security of cryptographic transformations on lattices is often reduced to the estimation of the minimum block size in the lattice reduction algorithm. For the expansion of small vectors, a reduction algorithm can be obtained for a given block size, the GSA model is often used, which uses the so-called Hermitian factor to predict the size of the vectors that the lattice reduction algorithm can obtain given the parameters. Asymptotic formulas have been developed to evaluate it in practice, but the question of their accuracy on cryptographic lattices has not been fully investigated. The work obtained estimates of the accuracy of the existing asymptotic estimates of the Hermite factor for lattices of sizes 120, 145, 170 for the classical BKZ algorithm. Research was conducted using the fpylll library. It was shown that the existing estimators are equivalent from a practical point of view and have a sufficiently small root mean square deviation from the true values. A formula was obtained that binds the root-mean-square error of approximation of the Hermit factor to the cryptographic parameters of lattices. The obtained results are useful for refining the security assessments of existing cryptographic transformations.
Abstract The magnetic behaviour of Cs2MYbF6 (M = Na, K, Rb) and Cs2NaYbBr6 has been studied in the temperature range between 3.5 and 251.3 K. The magnetic data are interpreted by means of a previously developed model in which the influence of the crystal field is theoretically described by the angular overlap model. The obtained values of the angular overlap parameter eσ(R) for the individual compounds are discussed and compared with each other. The energy values of the crystal-field levels of the 2F7/2 ground state are calculated.
A.V. Perfilyeva, Institute of Genetics and Physiology, CS MES RK, K.B. Bespalova
et al.
Tazy dogs appeared on the territory of Kazakhstan more than seven thousand years ago, and since then have become the national treasure of the Kazakh people. According to unofficial estimates from "KANSONAR", there are currently only about 3000 Tazy dogs left, but no more than 350 high-class dogs with pedigrees up to the fourth generation. The conservation and management of such valuable and rare breeds requires the study of their gene pool and systematic control of population genetic parameters. In Kazakhstan, as well as globally, genetic studies of dogs of the Tazy breed have not been conducted previously. The aim of this study was to analyse the parameters of genetic diversity for Tazy populations from different regions of Kazakhstan. Microsatellite analysis of 19 loci was conducted using Thermo Scientific Canine Genotypes Panel 1. The first locus, AHTk211, was selected from a list of 19 loci recommended by the International Society of Animal Genetics (ISAG). These included CXX279, REN169O18, INU055, REN54P11, INRA21, AHT137, REN169D01, AHTh260, AHTk2, and others. As a result, genetic profiles were obtained for Tazys from the southern, northern and eastern regions of Kazakhstan. The sufficiently high genetic variability and relatively low level of inbreeding quantitatively determined by polymorphic microsatellites in Tazys from the southern and to a lesser extent from the northern and eastern regions of Kazakhstan could be related to a large genetic variability in the foundation animals of this breed or to the metisation of the breed. The results obtained could potentially be used for the development of scientifically based breeding programmes for the management and breeding of this breed.
This paper presents a comprehensive analysis of two leading artificial intelligence (AI) systems – ChatGPT-4 from OpenAI and Bard from Google AI. It also provides an overview of the development of artificial intelligence in various fields and its impact on human daily life, especially in areas such as medicine, finance, public administration, etc. A detailed comparison of different versions of ChatGPT (GPT-3 and GPT-4) is carried out by discussing and analyzing their capabilities, improvements, and limitations. The article also discusses the integration of the Bard system with Google services, its unique functionality, and the latest updates. The purpose of the study is to compare the capabilities of ChatGP-4T and Bard AI systems, highlight their strengths and weaknesses, as well as their practical application. Comparative testing was conducted to evaluate the performance of each model (system) in various tasks, including solving a logical problem, writing an essay, analyzing followed by making suggestions for improving the website and writing HTML/CSS code for a web page. The results highlight the fact that, despite the recognized advantages of these models, their functional characteristics may sometimes be limited or not meet expectations when performing specific tasks, and the choice of system (model) will be adjusted depending on the needs of users.
Los Laboratorios de Paz han sido programas de cooperación al desarrollo por parte de actores internacionales como la Unión Europea en países con conflicto armado interno como Colombia. Concretamente, los diálogos de paz adelantados en el segundo decenio del siglo XXI entre el Gobierno Nacional y la guerrilla de las FARC permiten considerar la factibilidad de construir Laboratorios de Paz con la participación de la comunidad internacional en escenarios de postconflicto armado en Colombia. El caso del municipio de Viotá, Cundinamarca, el cual se encuentra en un proceso complejo de reconstrucción de su tejido social, representa la oportunidad de diseñar, formular e implementar un Laboratorio de Paz capaz de ser sostenible en una etapa de postconflicto en el país. Para ello, será fundamental contar con el apoyo de actores sociales e instancias de carácter local, regional, nacional e internacional en aras de que esta experiencia pueda ser replicada en otros municipios de Colombia.
The nonlaminar flow characteristics of Cs+ have been studied in an intense Cs+ gun. In the physical model established here, the effects of space charge and thermal initial velocity are taken into account, and a Maxwellian distribution of the thermal initial velocity and its cosine angular distribution of current density with respect to the normal line of the ion-emitting surface are assumed. Poisson’s equations are solved in terms of iteration to find a self-consistent solution, and the nonlaminar trajectories are traced; hence the phase diagram at the exit of the gun is obtained. The calculated results are compared with experimental data, and they are in good agreement.