Speaker: Isaia Belardinelli
7 Luglio 2026 | 14:15
DEIB, Sala Riunioni PT1 (Ed. 20A)
Per maggiori informazioni: Silvia Cascianelli | silvia.cascianelli@polimi.it
Sommario
Tuesday, July 7th, 2026 at 2:15 pm a new appointment of Data Science Seminars: Bioinformatics focus will take place in DEIB PT1 Meeting Room (Building 20A) organized by the Data Science for Bioinformatics group.The seminar will be held by Isaia Belardinelli, master's student in Computer Science, on the following subject: "Discovering ecological networks shaped by habitat conditions: a novel approach using customizable machine learning analyses on eDNA metabarcoding data".
Until recently, tracking ecosystem health required the direct observation of organisms—a costly and time-consuming process that struggled to capture complex ecological interactions. Environmental DNA (eDNA) metabarcoding has radically transformed this field. By sequencing DNA fragments directly from environmental samples, it allows the simultaneous detection of multiple organisms, making the monitoring of biodiversity, anthropogenic impacts, and the consequences of habitat health on human communities faster and more cost-effective than ever, via a routine sequencing process. However, current frameworks for reconstructing ecological communities from eDNA data require extensive spatial-temporal datasets and biased a-priori annotations of organism interactions. These constraints frequently restrict the method to large-scale projects and heavily limit the discovery of novel or site-specific ecological networks.
To address these challenges, we introduce a customizable pipeline that reconstructs ecological networks without a-priori annotations, adaptable to varying sample sizes across space and time. By integrating unsupervised machine learning with specific mathematical metrics, the pipeline translates biological and environmental data into weighted graphs. Within these networks, edges representing organism interactions are modulated by environmental parameters that influence species abundance. Additionally, a Python-based Graphical User Interface (GUI) was developed to facilitate data exploration, network construction, and the evaluation of how effectively the detected clusters reflect real ecosystems using calibrated validation metrics. To further scale the method, it is possible within the same GUI to download eDNA and environmental data from accredited online repositories. The proposed framework was evaluated using a marine eDNA metabarcoding dataset from three sites in the Puglia region, which differ in human presence and expected species variety. By applying community detection algorithms such as Louvain and Infomap, the pipeline captured distinct seasonal trends and identified ecological clusters correlating with habitat conditions. While further testing across diverse ecosystems and datasets is necessary to fully assess the framework's scalability and robustness, these preliminary results suggest that an unsupervised, data-driven approach can aid in uncovering biological interactions without a-priori annotation bias.