Alan Nafiiev

Public leaderboards such as the Therapeutics Data Commons (TDC) ADMET benchmark are widely treated as a ranking of state-of-the-art models. However, a high leaderboard position is only meaningful if the corresponding model can actually be reproduced and deployed by an independent researcher. In this work, we audit whether the top-ranked TDC ADMET models meet that bar. We assessed the top-ranked models of all 22 TDC ADMET leaderboards from the perspective of an end user with access only to the publicly released artifacts of each model─its publication, code repository, and installation instructions. For every end point, the top three models were screened with a unified protocol including an execution environment reproducibility check, a data-leakage assessment, verification of the hyperparameter-optimization procedure, and a reevaluation against the current leaderboard. Only three models (CaliciBoost, MapLight, and MapLight + GNN) passed all stages and reproduced their reported performance. The remaining models failed because of unavailable code, nonreproducible environments, runtime incompatibilities, or methodological flaws. We traced direct or indirect data leakage in the MiniMol, GradientBoost, and XGBoost models, and used deliberately overfitted variants of our own Mol2Vec-based models to show that tuning on the public test set─whether accidental or intentional─can substantially inflate both metrics and leaderboard rank. These results indicate that current TDC leaderboard positions cannot be read as a direct measure of model quality and practical applicability and emphasize the urgent need for better public ADMET benchmarks based on the hidden test sets, strict data set versioning and model submission with standardized inference environments.

Classical protein-ligand docking has been a cornerstone technique in computational drug discovery for decades but has reached an accuracy and performance plateau. Recently introduced Machine Learning (ML)-based docking methods offer a promising paradigm shift, but their practical adoption is hampered by accuracy-to-speed trade-offs, inadequate benchmarking standards, and questionable chemical validity of predicted poses. In this study, we introduce ArtiDock─an ML-based docking technique optimized for high-throughput virtual screening applications. To evaluate ArtiDock, we developed a dedicated performance and accuracy benchmark for pocket-specific rigid protein-ligand docking, which mimics realistic industrial drug discovery scenarios and is based on the novel PLINDER data set. We demonstrate that ArtiDock is 29-38% more accurate in comparison to leading open-source and commercial classical docking techniques such as AutoDock, Vina, and Glide, while providing a low computational cost. ArtiDock notably excels in challenging docking scenarios involving unbound protein structures and binding sites containing ions and structured water molecules. Additionally, we demonstrated competitive accuracy of our approach at considerably higher throughput compared to a wide range of AI docking and AI cofolding methods using the PoseX benchmark. Our results show that ArtiDock could be considered as a method of choice in high-throughput virtual screening scenarios.

Rigid docking remains the most popular method of predicting protein-protein interactions in cases when experimental 3D structures of the complexes are not available. The docking often relies on known unbound (Apo) protein structures, which may differ significantly from their bound (Holo) forms. Modern machine learning (ML) based conformational sampling techniques allow generating ensembles of functionally relevant protein structures, which may be closer to their Holo forms and thus could improve the outcomes of the classical rigid protein-protein docking. Here, we sampled conformations of the protein subunits in 30 complexes from the novel PINDER data set with two state-of-the-art ML-based techniques and evaluated their docking performance using several physics-based, data-based, and ML-based scoring functions. We showed that such conformational sampling rarely produces structures that are closer to the Holo conformations than the corresponding Apo ones. Moreover, even when such conformations are generated, none of the tested scoring functions were able to prioritize and rank them correctly. Our work highlights critical limitations in the current ML-enhanced rigid protein-protein docking workflows and emphasizes the need for new approaches that can better utilize the potential of modern techniques for conformational generation and scoring.

Motivation: Accurately identifying and prioritizing protein binding pockets is a foundational element of small-molecule drug discovery. Defining these known pockets currently relies on a laborious manual process of extracting key residue data from selected publications, reconciling inconsistent terminology, and independently computing volumetric representations. This manual curation to ensure biological relevance is time-consuming, error-prone, and represents a major bottleneck for efficient, high-throughput drug discovery. Results: We present a novel approach for the identification and prioritization of protein binding pockets for small molecules by combining geometric pocket detection with large language models (LLMs). Our method leverages Fpocket to generate candidate pockets, which are then validated against published experimental data extracted from research articles using LLM with a series of prompts fine-tuned to identify and extract residue-level information associated with experimentally confirmed binding sites. We developed a curated benchmark dataset of diverse proteins and associated literature to train and evaluate the LLM’s performance in paper relevance assessment and pocket extraction. Availability and implementation: The developed benchmark dataset and methodology are freely available at the GitHub repository (https://github.com/receptor-ai/LLM-benchmark-dataset) and Zenodo (DOI: 10.5281/zenodo.15798647).

The accurate prediction of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties is crucial for early-stage drug development, enabling the reduction of late-stage attrition and guiding compound prioritization. In recent years, machine learning models have emerged as powerful tools for ADMET prediction, leveraging diverse molecular representations ranging from handcrafted descriptors to graph neural networks and language model embeddings. Despite these advances, balancing predictive performance with computational efficiency remains a key challenge, particularly for high-throughput screening scenarios. Among unsupervised embedding methods, Mol2Vec has shown promise by capturing chemical substructure context analogously to word embeddings in natural language processing. However, its performance on comprehensive ADMET benchmarks has not been systematically assessed. In this work, we reimplement Mol2Vec with an expanded training corpus and higher embedding dimensionality, and evaluate its utility across 16 ADMET prediction tasks from the Therapeutics Data Commons (TDC). We show that while Mol2Vec embeddings alone are competitive, combining them with classical molecular descriptors and applying feature selection significantly improves performance. Our final MLP models with enhanced Mol2Vec embeddings achieved top-1 results in 10 of 16 benchmarks, outperforming all previously reported models on the TDC leaderboard in this regard, demonstrating that descriptor-enriched representations, paired even with relatively simple MLPs, can rival or exceed the performance of more complex models.

Rigid docking remains the most popular method of predicting protein-protein interactions in cases when experimental 3D structures of the complexes are not available. The docking often relies on known unbound (Apo) protein structures, which may differ significantly from their bound (Holo) forms. Modern machine learning (ML) based conformational sampling techniques allow generating ensembles of functionally relevant protein structures, which may be closer to their Holo forms and thus could improve the outcomes of the classical rigid protein-protein docking. Here, we sampled conformations of the protein subunits in 30 complexes from the novel PINDER dataset with two state-of-the-art ML-based techniques and evaluated their docking performance using several physics-based, data-based, and ML-based scoring functions. We showed that such conformational sampling rarely produces structures that are closer to the Holo conformations than the corresponding Apo ones. Moreover, even when such conformations are generated, none of the tested scoring functions were able to prioritize and rank them correctly. Our work highlights critical limitations in the current ML-enhanced rigid protein-protein docking workflows and emphasizes the need for new approaches that can better utilize the potential of modern techniques for conformational generation and scoring.

We present a novel approach for the identification and prioritization of protein binding pockets for small molecules by combining geometric pocket detection with Large Language Models (LLMs). Our method leverages Fpocket to generate candidate pockets, which are then validated against published experimental data extracted from research articles using LLM with a series of prompts fine-tuned to identify and extract residue-level information associated with experimentally confirmed binding sites. We developed a curated benchmark dataset of diverse proteins and associated literature to train and evaluate the LLM’s performance in paper relevance assessment and pocket extraction. The extracted information is then mapped onto protein structures and used to filter and merge the geometry-based predictions, generating a refined volumetric representation of biologically relevant pockets. This hybrid pipeline offers an efficient, accurate and automated method for identifying functional binding pockets, addressing a significant bottleneck in the high-throughput drug discovery workflows. The developed benchmark dataset and methodology are freely available at https://github.com/MelnychenkoM/LLM-benchmark-dataset .

Subject matter. This article presents the development and architecture of an automated program complex designed to identify and analyze malicious executable files using a classifier based on a multiple kernel support vector machine (SVM). Goal. The aim of the work is to create an automated system that enhances the accuracy and efficiency of malware detection by combining static and dynamic analysis into a single framework capable of processing large volumes of data with optimal time expenditure. Tasks. To achieve this goal, tasks were carried out that included developing a program complex that automates the collection of static and dynamic data from executable files using tools like IDA Pro, IDAPython, and Drakvuf; integrating a multiple kernel SVM classifier to analyze the collected heterogeneous data; validating the system's effectiveness based on a substantial dataset containing 1,389 executable samples; and demonstrating the system's scalability and practical applicability in real-world conditions. Methods. The methods involved a hybrid approach that combines static analysis – extracting byte code, disassembled instructions, and control flow graphs using IDA Pro and IDAPython – with dynamic analysis, which entails monitoring real-time behavior using Drakvuf. The multiple kernel SVM classifier integrates different data representations using various kernels, allowing for both linear and nonlinear relationships to be considered in the classification process. Results. The results of the study show that the system achieves a high level of accuracy and completeness, as evidenced by key performance metrics such as an F-score of 0.93 and ROC AUC and PR AUC values. The automated program complex reduces the analysis time of a single file from an average of 11 minutes to approximately 5 minutes, effectively doubling the throughput compared to previous methods. This significant reduction in processing time is critically important for deployment in environments where rapid and accurate malware detection is necessary. Furthermore, the system's scalability allows for efficient processing of large data volumes, making it suitable for real-world applications. Conclusions. In conclusion, the automated program complex developed in this study demonstrates significant improvements in the accuracy and efficiency of malware detection. By integrating multiple kernel SVM classification with static and dynamic analysis, the system shows potential for real-time malware detection and analysis. Its scalability and practical applicability indicate that it could become an important tool in combating modern cyber threats, providing organizations with an effective means to enhance their cybersecurity.

Classical protein-ligand docking has been a cornerstone technique in computational drug discovery for decades, but has reached an accuracy and performance plateau. Recently introduced Machine Learning (ML) based docking methods offer a promising paradigm shift, but their practical adoption is hampered by accuracy-to-speed trade-offs, inadequate benchmarking standards, and questionable chemical validity of predicted poses. In this study, we introduce ArtiDock - an ML-based docking technique optimized for high-throughput virtual screening applications. To evaluate ArtiDock, we developed a dedicated performance and accuracy benchmark for pocket-specific rigid protein-ligand docking, which mimics realistic industrial drug discovery scenarios and is based on the novel PLINDER dataset. We demonstrate that ArtiDock is 29-38% more accurate in comparison to leading open-source and commercial classical docking techniques such as AutoDock, Vina, and Glide, while providing a low computational cost. ArtiDock notably excels in challenging docking scenarios involving unbound protein structures and binding sites containing ions and structured water molecules. Our results show that ArtiDock could be considered as a method of choice in high-throughput virtual screening scenarios.

Every year, malware authors create more and more sophisticated and clever malware that can harm our computers. Traditional methods, which are based on searching for program signatures are no longer effective in solving the problem of malware detection. It is being replaced by automated file analysis, which is a more promising approach to detecting suspicious files. Machine learning methods are increasingly used to detect such malware programs. However, such solutions may require a lot of computing resources to perform their operations. Therefore, the task of creating an optimal machine learning model in terms of learning speed and malware detection accuracy arises. In addition, usually one method of data representation is not sufficient to detect malicious features of files. Therefore, this paper will describe two different methods: one method is based on the binary information of the file, the other one is based on disassembled code of executable files. The purpose of this work is to improve the efficiency of malware detection by optimising feature extraction methods and applying machine learning. The main tasks of the study include: extracting features from exe files, creating several machine learning models and comparing them to determine the most effective one. The dataset used in this study has been collected from various online sources and consists of 12824 executable files in .exe format, of which 11844 files are malicious and 980 are benign. This paper presents recommended methods of feature extraction and input data generation for machine learning models based on the support vector machine algorithm. These methods allow to find the best way to process the features describing a malicious file. Six machine learning models, each of which performed well in terms of F-score, precision, and recall metrics, were created. The model that was created based on the binary type of data representation showed the highest results for all metrics. З кожним роком автори шкідливого програмного забезпечення створюють все більш досконалі та хитромудрі шкідливі програми, які можуть завдати шкоди нашим комп’ютерам. Традиційні методи, які ґрунтуються на пошуку сигнатур програм, перестають бути ефективними для вирішення проблеми детекції шкідливого програмного забезпечення. На зміну приходить автоматизація аналізу файлів, яка є більш перспективним підходом для виявлення підозрілих файлів. Для виявлення таких програм все частіше використовують методи машинного навчання. Однак для виконання своїх операцій такі рішення можуть потребувати багато обчислювальних ресурсів. Тому виникає задача створення оптимальної моделі машинного навчання з погляду швидкості навчання і точності детекції шкідливого програмного забезпечення. Крім того, зазвичай одного методу представлення даних недостатньо для якісного виявлення шкідливих ознак файлів. Тому в цій роботі буде описано два різні методи: один підхід ґрунтується на бінарній інформації файлу, другий полягає у використанні трасувальних інструкцій. Мета цієї роботи – підвищення ефективності виявлення шкідливого програмного забезпечення шляхом оптимізації методів вилучення ознак та застосування машинного навчання. Основні задачі дослідження включають: вилучення ознак з exe. файлів, створення кількох моделей машинного навчання та їх порівняння для визначення найефективнішої моделі. Використаний у цьому дослідженні набір даних був зібраний з різних інтернет-джерел та складається з 12824 виконуваних файлів у форматі .exe, з яких 11844 файлів є шкідливими, а 980 – доброякісними. У статті представлено рекомендовані методи вилучення ознак та генерації вхідних даних для моделей машинного навчання на основі алгоритму машини опорних векторів. Ці методи дозволяють знайти найкращий шлях для обробки ознак, що описують шкідливий файл. Було створено шість моделей машинного навчання, кожна з яких показала високі показники метрик F-score, precision та recall. Модель, яка була створена на основі бінарного типу представлення даних, показала найвищі результати по всіх метриках.