Kil To Chong

Persistent viruses like Influenza, HIV, and Coronavirus exemplify the challenge of viral escape, significantly hindering the development of long-lasting vaccines and effective treatments. This study leverages a Long Short-Term Memory (LSTM) based deep learning architecture to analyze an extensive dataset of over 3.1 million unique viral spike protein sequences, with SARS-CoV-2 serving as the primary example. Our model, Escape Elite Network (EEN) outperforms existing methods in detecting escape mutations across diverse datasets. In computational (Validation) and wet lab (Baum and Greaney) datasets, EEN achieved AUC scores of 0.949, 0.868, and 0.762, respectively, each with p-values less than $1\times 10^{-5}$, demonstrating high statistical significance. Specifically, EEN significantly outperformed the competitor models NPVE and SEN across all datasets. For the Greaney dataset, EEN achieved a 19.2% improvement over NPVE and a 5.7% improvement over SEN. Similarly, for the Baum dataset, EEN showed a 1.8% improvement over NPVE and a 12.8% improvement over SEN. This tool proactively predicts viral escape mutations, guiding the development of more effective vaccines and therapeutics. It identifies high-risk mutations before experimental validation, offering advantages in combating SARS-CoV-2 and potentially other viruses like HIV, Influenza, and African Swine Fever, with broad applications in virology and epidemiology.

Janus kinase 2 (JAK2) is an important therapeutic target for various inflammatory diseases, cancers, and rheumatoid arthritis. Therefore, inhibiting JAK2 has become a promising approach for treating these conditions. In this study, molecular descriptors such as Morgan fingerprints, Molecular Access System (MACCS), and PaDEL were calculated and used to develop machine-learning models. Among these models, CatBoost combined with Morgan fingerprints performed the best, achieving an accuracy of 0.94 on the test dataset. This CatBoost model was then used to screen the Korean Chemical Databank (KCB) to identify the most potent JAK2 inhibitors. Computational analyses, including density functional theory (DFT), molecular docking, and molecular dynamics simulations, were carried out to evaluate the performance of the top-ranked molecules. Finally, four compounds were selected for experimental testing, and the results showed that their IC₅₀ values were less than 10 μM. The integration of AI-driven modeling with experimental validation provides a promising strategy for personalized medicine, enabling the development of more precise and effective kinase-targeted therapies while reducing the time and cost required to bring new drugs to clinical trials.

Highway accidents remain a critical public safety challenge, where delayed response times intensify injury severity and fatalities. Traditional systems relying on manual reporting suffer from inefficiencies and critical delays during emergencies. This paper presents an autonomous drone based highway incident management system that integrates advanced aerial technology with Vision-Language Models (VLMs) and Large Language Models (LLMs) to enable rapid detection, analysis, and response. The framework employs a DJI Matrice 30T drone equipped with a dock station for autonomous deployment for incident analysis by yielding incident coordinates provided by authorities. Upon arrival, real-time video is analyzed using an optimized fine-tuned YOLOv12n model selected after rigorous comparison of nine YOLO variants to detect incidents such as collisions and fires with high precision. To generate actionable intelligence, high-confidence video frames are processed using a hybrid vision-language pipeline comprising of LLaVA-OneVision-Qwen2 and GPT-4o API. This combination yields detailed natural language descriptions and structured summaries of incident scenes, offering enhanced contextual awareness of the incident site. By combining autonomous drone navigation, real-time object detection, VLM-based scene interpretation, and LLM-driven summarization, this end-to-end solution enhances situational awareness and decision-making for emergency responders. These results demonstrate the practical value of integrating multi-modal AI with UAVs in critical safety domains, laying the groundwork for scalable, intelligent transportation infrastructure and next-generation emergency response systems.

Kinase is an enzyme responsible for cell signaling and other complex processes. Mutations or changes in kinase can cause cancer and other diseases in humans, including leukemia, neuroblastomas, glioblastomas, and more. Considering these concerns, inhibiting overexpressed or dysregulated kinases through small drug molecules is very important. In the past, many machine learning and deep learning approaches have been used to inhibit unregulated kinase enzymes. In this work, we employ a Graph Neural Network (GNN) to predict the inhibition activities of kinases. A separate Graph Convolution Network (GCN) and combined Graph Convolution and Graph Attention Network (GCN_GAT) are developed and trained on two large datasets (Kinase Datasets 1 and 2) consisting of small drug molecules against the targeted kinase using 10-fold cross-validation. Furthermore, a wide range of molecules are used as independent datasets on which the performance of the models is evaluated. On both independent kinase datasets, our model combining GCN and GAT provides the best evaluation and outperforms previous models in terms of accuracy, Matthews Correlation Coefficient (MCC), sensitivity, specificity, and precision. On the independent Kinase Dataset 1, the values of accuracy, MCC, sensitivity, specificity, and precision are 0.96, 0.89, 0.90, 0.98, and 0.91, respectively. Similarly, the performance of our model combining GCN and GAT on the independent Kinase Dataset 2 is 0.97, 0.90, 0.91, 0.99, and 0.92 in terms of accuracy, MCC, sensitivity, specificity, and precision, respectively.

Nanobodies offer significant therapeutic potential due to their small size, stability, and versatility. Although advancements in computational protein design have made designing de novo nanobodies increasingly feasible, there are limited tools specifically tailored for this purpose. Rosetta with its specialized protocols, is a prominent tool for nanobody design but is limited by a high false-negative rate, necessitating extensive high-throughput screening. This results in increased costs, time, and labor due to the need for large-scale experimentation and detailed structural analysis. To address current challenges in nanobody design, we introduce NanoBinder, an interpretable machine learning model that predicts nanobody-antigen binding using Rosetta energy scores. NanoBinder utilizes a Random Forest model trained on experimentally validated complexes and can be seamlessly integrated into the Rosetta software. It employs SHAP summary plots for interpretability, which helps identify key features influencing binding interactions. Experimentally validated on forty-nine diverse nanobodies, NanoBinder accurately predicts non-binders and shows reasonable performance in identifying binders. This approach significantly enhances predictive accuracy, reduces the need for extensive experimental assays, and accelerates nanobody development, thereby offering a powerful tool to mitigate the costs, time, and labor associated with high-throughput screening. Scientific contribution This study introduces NanoBinder, a machine learning framework for predicting nanobody-antigen binding using Rosetta-derived energy features. Through rigorous experimental validation across diverse nanobody sets, NanoBinder enhances nanobody screening workflows by reducing false positives and minimizing reliance on extensive wet-lab assays. The approach bridges the gap between physics-based modeling and data-driven prediction in nanobody design.

Background/Objectives: Pomegranate (Punica granatum) peel, often discarded as waste, contains abundant bioactive compounds such as polyphenols, vitamins, flavonoids, tannins, anthocyanins, and many more. This contributes to remarkable bioactivities, including anticancer, anti-inflammatory, antioxidant, antibacterial, and antifungal properties. Pancreatic cancer is a deadly cancer with a 9% survival rate. Its aggressiveness, invasiveness, quick metastasis, and poor prognosis significantly decrease the survival rate. Thus, we aim to explore pomegranate peel as a possible alternative medication for treating pancreatic cancer through virtual methods. Methods: Firstly, bioactive compounds were collected from multiple databases and screened for oral bioavailability (OB) ≥ 0.3 and drug likeness (DL) ≥ 0.18 scores. Simultaneously, network pharmacology was employed to extract the most probable targets for pancreatic cancer. Further computational analyses were performed, including molecular docking, molecular dynamics simulation, and in silico pharmacokinetics evaluation. Results: Consequently, the top 10 key targets from network analysis were AKT1, IL6, TNF, SRC, STAT3, EGFR, BCL2, HSP90AA1, HIF1A, and PTGS2. However, only AKT1, EGFR, BCL2, HSP90AA1, and PTGS2 exhibited strong binding affinities with pomegranate compounds, which are significantly declared in affected cells to enhance cancer progression. Outcomes from molecular dynamics simulations, particularly RMSD, RMSF, hydrogen bonding, and radius of gyration (Rg), confirmed stable interactions between 1-O-Galloyl-beta-D-glucose, epicatechin, phloridzin, and epicatechin gallate with respective target proteins. Conclusions: This suggests that pomegranate peels hold anticancer bioactive compounds for treating pancreatic cancer. Surprisingly, most compounds adhere to Lipinski’s and Pfizer’s rules and display no toxicity. However, as this study relies entirely on computational methods, experimental validation is necessary to confirm these findings and assess real-world efficacy and potential side effects.