IF3211 Project Gallery – Domain-Specific Computing

Breast cancer is one of the most common types of cancer and is the leading cause of cancer death in women worldwide. One of the main factors contributing to breast cancer is genetic mutations in tumor suppressor genes. Early detection of these genetic mutations is crucial for the prevention and management of breast cancer. This study proposes the development of a Deep Learning-based machine learning model to detect genetic mutations in DNA sequences, specifically in the BRCA1 and BRCA2 genes, with the aim of providing fast and accurate detection results. The applied models include Temporal Convolutional Network (TCN), One-dimensional Convolutional Neural Network (1D-CNN), and Recurrent Neural Network (RNN) BiLSTM, using sequential labeling techniques to identify the type and index of mutations in each nucleotide. The DNA sequence data used comes from breast cancer patient samples from the Catalogue of Somatic Mutation in Cancer (COSMIC). The test results show that TCN has the best performance with an F1-score of 0.8602, outperforming ID-CNN (0.8325) and BiLSTM (0.8325), and showing a shorter detection time. Analysis of the detected mutations shows a predominance of C→A and C→T mutations, consistent with the literature on human genomic mutations as driver mutations in many tumor suppressor genes. This research contributes to the use of machine learning for bioinformatics analysis, particularly in the automated and rapid detection of genetic mutations, which is expected to support early diagnosis and more effective management of breast cancer.

The rhizosphere is a biologically active soil zone strongly influenced by root activity, where microbial communities play a crucial role in nutrient cycling and ecosystem health. This study aimed to examine how urban wastewater exposure affects the structure and diversity of bacterial communities in the rhizosphere, identify dominant or significantly altered taxa, and identify potential microbial biomarkers that play a role in the response to environmental stress. Analysis was conducted on 27 metagenome samples (BioProject PRJNA1229183) using 16S rRNA gene sequencing (V3–V4) and analyzed with QIIME2, DADA2, and PICRUSt2. The results showed differences in microbial community structure between samples based on β-diversity analysis, with the dominance of the phyla Proteobacteria, Actinobacteriota, and Firmicutes. Several genera such as Pseudomonas, Geobacter, Hydrogenophaga, and Acidovorax were identified as having important roles in nitrogen fixation and bioremediation. Genetic function prediction revealed the dominance of rpoE and ABC-type transporter genes associated with membrane stress responses, which have potential as molecular biomarkers. This study provides initial insights into rhizosphere microbial adaptation to urban wastewater stress and identifies taxa and functional genes relevant for soil ecosystem monitoring and restoration strategies in urban environments.

Evolution is a concept in biology that explains the changes in the characteristics of an organism and is passed down to the next generation. Living organisms use evolution to adapt to their environment, thereby increasing their lifespan. Mutations represent changes in genetic material. To study this, a simulation will be designed to visualize evolution. Bacteria will be used as the primary research subject due to their unicellular nature and relatively simple genetic structure. The simulation specifically focuses on the evolution of antibiotic resistance in bacterial populations, where selection pressure from antibiotic exposure triggers mutations that enable bacterial survival. The simulation model uses a genetic algorithm to represent the processes of mutation, natural selection, and bacterial reproduction in environments containing varying concentrations of antibiotics. This evolution is then visualized to observe patterns of resistance development and the dynamics of bacterial evolution over time. The simulation results are expected to provide a better understanding of the mechanisms of antibiotic resistance evolution and the factors influencing its rate of development at the population level.

Orangutan conservation faces critical challenges due to habitat fragmentation and genetic bottlenecks across three species: Sumatran (Pongo abelii), Bornean (Pongo pygmaeus), and Tapanuli (Pongo tapanuliensis). Traditional conservation approaches lack integration of genomic diversity with spatial distribution data. This paper presents PhyloGeoVis, a computational framework that combines phylogenomic analysis with interactive geospatial visualization to prioritize orangutan conservation efforts. Our approach employs multiple sequence alignment, maximum likelihood phylogenetic reconstruction, population viability analysis, and GIS integration to analyze genomic sequences from NCBI GenBank databases. The system provides conservation decision support through interactive visualization of genetic diversity patterns, identification of genomic regions under selection pressure, and extinction risk assessment. Performance evaluation demonstrates computational efficiency and biological relevance for conservation practitioners.

Gene Regulatory Networks (GRNs) play a crucial role in understanding complex molecular mechanisms underlying plant growth and development. This study presents a novel computational approach for modeling GRNs in Arabidopsis thaliana using Graph Neural Networks (GNNs). We utilized gene co-expression data from the ATTED-II database to construct correlation graphs where nodes represent genes and edges represent significant correlations between gene pairs. The GNN architecture was implemented using PyTorch to learn complex regulatory patterns and generate low-dimensional vector embeddings for each gene. These embeddings were then used to identify hub genes and construct the regulatory network structure. To evaluate the constructed GRN, we employed GO enrichment analysis by comparing positive and negative gene sets based on their regulatory distance from identified master regulator genes. Our results demonstrate that PSBO1 emerges as a key master regulator gene, with GO enrichment analysis showing significantly lower p-values and FDR rates for genes directly regulated by PSBO1 compared to distantly regulated genes (difference factor of less than 10^-6). This validation confirms PSBO1’s central role in photosynthesis regulation, consistent with biological literature. Our approach successfully demonstrates the effectiveness of GNNs in inferring regulatory relationships from co-expression data and identifying biologically meaningful hub genes in plant regulatory networks.

Mammalian species have evolved diverse physiological and morphological adaptations to survive in extreme environments, mainly classified to be aquatic and marine, terrestrial, and non-aquatic caves and subterranean. Understanding these adaptation patterns is crucial for evolutionary biology research and conservation efforts. This study presents a machine learning approach to classify mammalian species into their primary habitats based on biological traits. We developed supervised classification models using a comprehensive dataset of 1456 mammalian species with 31 biological features including life history traits, morphological traits, reproductive traits, ecological traits, and behavioral traits. Random Forest, combined with Multi Output Classifier was evaluated using holdout testing. The model achieved high classification accuracy and identified key biological features most indicative of habitat adaptation. Results demonstrate the effectiveness of machine learning in revealing complex species-environment relationships and provide insights for predicting mammalian responses to environmental changes.

This study developed an agent-based simulation model (ABM) to study the dynamics of bird species richness and population in the Galapagos Islands. The simulation environment was constructed from GIS data of the Galapagos Islands, which was processed using GeoPandas and Shapely into a spatial grid containing habitat types. Each agent represented a single bird with biological attributes such as reproductive rate, mortality, energy range, habitat preference, and dispersal ability, normalized per weekly step. The core processes of random immigration from the mainland, dispersal, energy acquisition, mortality, and reproduction were run at discrete timescales over several simulation years. Output data included total population trends per species, species richness per island (Species Area Relationship and Species Isolation Relationship), and final distribution maps of individuals over the habitat map. Simulation results showed that island size, habitat type, and isolation distance synergistically influenced variations in species richness, consistent with island biogeography theory. Validation using the Jaccard Index and Root Mean Square Error (RMSE) against field population estimation data confirmed the model's ability to replicate population composition and numbers with varying degrees of accuracy across islands. This model provides a quantitative tool for understanding ecological patterns in island ecosystems and supports data-driven conservation planning.