1081. |
1082. |
1083. Graph Convolutional Networks for Predicting Cancer Outcomes and Stage : a focus on cGAS-STING pathway activationMateo Sokač, Borna Skračić, Danijel Kučak, Leo Mršić, 2024, original scientific article Abstract: The study presented in this paper evaluated gene expression profiles from The Cancer Genome Atlas (TCGA). To reduce complexity, we focused on genes in the cGAS–STING pathway, crucial for cytosolic DNA detection and immune response. The study analyzes three clinical variables: disease-specific survival (DSS), overall survival (OS), and tumor stage. To effectively utilize the high-dimensional gene expression data, we needed to find a way to project these data meaningfully. Since gene pathways can be represented as graphs, a novel method of presenting genomics data using graph data structure was employed, rather than the conventional tabular format. To leverage the gene expression data represented as graphs, we utilized a graph convolutional network (GCN) machine learning model in conjunction with the genetic algorithm optimization technique. This allowed for obtaining an optimal graph representation topology and capturing important activations within the pathway for each use case, enabling a more insightful analysis of the cGAS–STING pathway and its activations across different cancer types and clinical variables. To tackle the problem of unexplainable AI, graph visualization alongside the integrated gradients method was employed to explain the GCN model’s decision-making process, identifying key nodes (genes) in the cGAS–STING pathway. This approach revealed distinct molecular mechanisms, enhancing interpretability. This study demonstrates the potential of GCNs combined with explainable AI to analyze gene expression, providing insights into cancer progression. Further research with more data is needed to validate these findings.
Keywords: cGAS–STING, graph-convolutional-network, graphs, cancer, pan-cancer, machine learning, NGS
Published in DiRROS: 09.09.2025; Views: 304; Downloads: 148
 Full text (2,05 MB)
This document has many files! More... |
1084. Structural and interfacial characterization of a photocatalytic titanium MOF-phosphate glass compositeCelia Castillo-Blas, Montaña J. García, Ashleigh M. Chester, Matjaž Mazaj, Shaoliang Guan, Georgina P. Robertson, Ayano Kono, James M. A. Steele, Luis León-Alcaide, Bruno Poletto Rodrigues, Andraž Krajnc, 2025, original scientific article Published in DiRROS: 09.09.2025; Views: 290; Downloads: 120
Full text (1,62 MB)
This document has many files! More... |
1085. |
1086. Charge density study of two-electron four-center bonding in a dimer of tetracyanoethylene radical anions as a benchmark for two-electron multicenter bondingMiha Virant, Petar Štrbac, Anna Krawczuk, Valentina Milašinović, Petra Stanić, Matic Lozinšek, Krešimir Molčanov, 2024, original scientific article Abstract: The dimer of the tetracyanoethylene (TCNE) radical anions represents the simplest and the best studied case of two-electron multicenter covalent bonding (2e/mc or pancake bonding). The model compound, N-methylpyridinium salt of TCNE•–, is diamagnetic, meaning that the electrons in two contiguous radicals are paired and occupy a HOMO orbital which spans two TCNE•– radicals. Charge density in this system is studied as a benchmark for comparison of charge densities in other pancake-bonded radical systems. Two electrons from two contiguous radicals indeed form a bonding electron pair, which is distributed between two central ethylene groups in the dimer, i.e., between four carbon atoms. The topology of electron density reveals two bond critical points between the central ethylene groups in the dimer, with maximum electron density of 0.185 e Å–3; the corresponding theoretical value is 0.118 e Å–3.
Published in DiRROS: 09.09.2025; Views: 275; Downloads: 57
Full text (4,85 MB) |
1087. |
1088. |
1089. Automated grading through contrastive learning : a gradient analysis and feature ablation approachMateo Sokač, Mario Fabijanić, Igor Mekterović, Leo Mršić, 2025, original scientific article Abstract: As programming education becomes increasingly complex, grading student code has become a challenging task. Traditional methods, such as dynamic and static analysis, offer foundational approaches but often fail to provide granular insights, leading to inconsistencies in grading and feedback. This study addresses the limitations of these methods by integrating contrastive learning with explainable AI techniques to assess SQL code submissions. We employed contrastive learning to differentiate between student and correct SQL solutions, projecting them into a high-dimensional latent space, and used the Frobenius norm to measure the distance between these representations. This distance was used to predict the percentage of points deducted from each student’s solution. To enhance interpretability, we implemented feature ablation and integrated gradients, which provide insights into the specific tokens in student code that impact the grading outcomes. Our findings indicate that this approach improves the accuracy, consistency, and transparency of automated grading, aligning more closely with human grading standards. The results suggest that this framework could be a valuable tool for automated programming assessment systems, offering clear, actionable feedback and making machine learning models in educational contexts more interpretable and effective.
Keywords: automated programming assessment systems (APASs), contrastive learning, explainable AI, feature ablation, integrated gradients, machine learning in education, natural language processing (NLP)
Published in DiRROS: 09.09.2025; Views: 300; Downloads: 149
Full text (4,12 MB)
This document has many files! More... |
1090. |
