Page 149 - SRPSKO DRUŠTVO ISTRAŽIVAČA RAKA

SDIRSACR Oncology Insights

where the device would display the probability that the tissue is affected by a tumor.
Material and Methods: The study included data from 140 patients, among whom were patients with healthy rectums.
In 130 patients, a tumor localized entirely in the rectum was diagnosed, while 10 patients with healthy rectums were
included to balance the dataset. Only those slices depicting the rectal region were selected, resulting in approximately
3,600 images suitable for analysis. A set of the most relevant features was extracted from the images, and a table
was created for training and testing multiple ML algorithms. The algorithms used were logistic regression (LR), linear
discriminant analysis (LDA), support vector machines (SVM), classification and regression trees (CART), naive Bayes
(NB), and k-nearest neighbors (KNN). In addition to classical ML algorithms, neural networks were trained to compare
performance and explore the potential of deep learning in tumor detection. Models were trained on the same feature
set with a training and testing split. Evaluation focused particularly on sensitivity and specificity parameters, which are
critical in medical diagnostics.
Results: The best result on the available dataset was achieved using the SVM algorithm, which reached over 80%
accuracy in tumor area detection. Neural networks demonstrated potential for higher sensitivity, with a need for
further model architecture tuning to improve specificity.
Conclusions: This approach can contribute to more efficient diagnostics, save resources and time for physicians, as well
as enabling more precise therapy planning and a personalized patient approach. Furthermore, potential integration of
the developed algorithms into medical equipment could improve automatic and rapid tumor detection during routine
diagnostic procedures.

Acknowledgments and funding: MM is supported by the Horizon Europe STEPUPIORS Project (HORIZON-WIDERA-
2021-ACCESS-03, European Commission, Agreement No. 101079217) and the Ministry of Science, Technological
Development and Innovation of the Republic of Serbia (Agreement No. 451-03-136/2025-03/200043).

P57

Bioinformatic evaluation of SMTN promoter-driven transcripts overexpressed in gastrointestinal cancers

Teodor Skendžić, Dunja Pavlović, Aleksandra Nikolić

Group for Gene Regulation in Cancer, Institute of Molecular Genetic and Genetic Engineering, University of Belgrade, Belgrade,
Serbia

Keywords: RNA, micropeptides, gastrointestinal neoplasms

Background: Recent evidence suggests that some non-canonical transcripts harbour small open reading frames
(sORFs) that encode microproteins with potential functional significance. Smoothelin, encoded by the SMTN gene,
is a cytoskeletal protein composed of 915 amino acids, primarily expressed in differentiated smooth muscle cells.
Transcripts SMTN-206 (ENST00000422839) and SMTN-209 (ENST00000432777) code for proteins 37 and 91 amino
acids long, respectively. Recent pan-cancer transcriptome analysis has revealed that the activity of the promoter
driving their expression is significantly increased in gastrointestinal tumours.
Material and Methods: Expression of SMTN-206 and SMTN-209 in tumor and non-tumor tissue was investigated using
TCGA and GTEx datasets via the USCS Xena Browser. Sequences of SMTN-206 and SMTN-209 were retrieved from the
Ensembl GRCh38 genome browser in FASTA format. Analyses included predictions of transcript localization, secondary
structure, and interactions with miRNA. Additionally, sORF detection and microprotein localization were performed for
SMTN-206. Ribosome profiling (RiboSeq) data were obtained from the GSE269371 dataset from NCBI Gene Expression
Omnibus and used for estimating ribosome occupancy in CaCo2 and HCEC-1CT cell lines.
Results: Expression data revealed expression of SMTN-206 and SMTN-209 in colon, rectum and stomach tumors,
suggesting tissue-specific promoter utilization. SMTN-206 demonstrated significant differential expression between
tumor tissue and healthy gut mucosa warranting its prioritization for further analysis. MicroRNA interaction predictions
indicated associations with miRNAs involved in tumor suppression and immune regulation. sORFs detection confirmed
a translated region located between nucleotides 456-566 producing microprotein with extracellular localization.
RiboSeq data confirmed differential ribosome occupancy between tumor-derived CaCo2 and non-tumor HCEC-1CT cell
lines, suggesting increased translation in tumor cells.
Conclusions: Given its tumor-specific expression, increased translation in tumor cells, and interactions with cancer-
relevant miRNAs, SMTN-206 emerges as a promising biomarker candidate in GI tumors. The encoded microprotein
warrants further investigation due to its significant structural divergence from the canonical protein and its potential

134