On Monday, 31/8/2026 at 15.00, Violetta Gkika, a graduate student of the program “Data Science and Information Technologies, Biomedical Data Science – Bioinformatics” will present their MSc thesis titled:

Exploring Epithelial Heterogeneity in Pancreatic Ductal Adenocarcinoma Using the MLscAN Computational Framework

Malignant epithelial cells in pancreatic ductal adenocarcinoma (PDAC) exhibit a continuous range of phenotypes that appear to connect the traditional “classical” and “basal” transcriptional subtypes, suggesting high cellular plasticity rather than fixed binary categories. To explore this spectrum without forcing static labels based on gene markers alone, this thesis employs MLscAN (Machine Learning for single-cell ANalytics), an unbiased, unsupervised, probabilistic, and transition-aware computational framework developed by Prof. Manolakos’s team at NKUA. We applied the analysis to a high-resolution dataset of 106.110 malignant cells from 50 patients. Starting from a batch-integrated embedding that separates common epithelial features from patient-specific variation, the probabilistic approach robustly identifies the underlying transcriptional cancer states and their interactions, driven entirely by the data. This analysis was done in collaboration with Dr. Kalfakakou from Prof. Tsirigos’s lab at NYU Medical Center, who provided the dataset analyzed [1].

Unsupervised state inference resolved the epithelial landscape into five main transcriptional states. State-to-state trajectory modeling revealed two distinct evolutionary routes connecting the classical and basal programs. The analysis shows that both transitional paths start with a decline in classical identity markers like GATA6 and HNF4A, but then diverge. One path features late activation of the ∆Np63/TP63 squamous commitment cascade combined with cell proliferation. The other shifts toward a basal phenotype via a different, non-proliferative effector program that avoids the master-switch cascade altogether. By showing that the classical-to-basal transition involves at least two separate biological processes, our results advance the discussion of a multi-routed evolutionary model in tumor biology. Moreover, this study underscores the value of the MLscAN framework, illustrating how probabilistic, explicit modeling of state transitions can reveal complex evolutionary patterns often hidden by standard clustering methods.

[1] D. Kalfakakou, A. Tsirigos, et al. “Clonal Heterogeneity in Human Pancreatic Ductal Adenocarcinoma.” bioRxiv, preprint, 2025. DOI: 10.1101/2025.02.11.637729. Available: https://www.biorxiv.org/content/10.1101/2025.02.11.637729v1

Examination Committee:

Prof. Em. Elias S. Manolakos (advisor), Dept. of Informatics and Telecommunications, NKUA

Dr. Kostas Vekrellis, Research Director, Biomedical Research Foundation of the Academy of Athens

Prof. Dimitrios J. Stravopodis, Assoc. Professor of Biology, Department of Biology, NKUA

Zoom link

Elias Manolakos is inviting you to a scheduled Zoom meeting.

Join Zoom Meeting

https://us02web.zoom.us/j/87187596046?pwd=B6FRWRgaDeMWvr7cbk18nTeyjLFSjs.1

Meeting chat link

https://us02web.zoom.us/launch/jc/87187596046

Meeting ID: 871 8759 6046

Passcode: 658586

—

One tap mobile

+302111984488,,87187596046#,,,,*658586# Greece

+302311180599,,87187596046#,,,,*658586# Greece

—

Join by SIP

• 87187596046@zoomcrc.com

Join instructions

https://us02web.zoom.us/meetings/87187596046/invitations?signature=EJWK28jpQIl47vJ-dN2lUJQiPWPINT-hM-B_OqePkmU