REA Press Null REA Press3042-31203042-3120 REA Press https://doi.org/10.22105/ahse.v3i3.67 Research Article Tumor type classification, Somatic mutations, Transformer model, Deep learning, TCGA, Precision oncology Tumor Type Classification from Somatic Mutations Using a Transformer-Based ModelTumor Type Classification from Somatic Mutations Using a Transformer-Based Model Sanaei Saeid Faculty of Electrical and Computer Engineering, Mazandaran University of Science and Technology, Behshahr, Iran. Javanmard Alitappeh Reza Faculty of Electrical and Computer Engineering, Mazandaran University of Science and Technology, Behshahr, Iran. 06 2026 10 06 2026 3 3 © 2026 REA Press 2026

This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

 Tumor Type Classification from Somatic Mutations Using a Transformer-Based Model

Somatic mutation profiles are sparse and variable in length, which makes tumor-type classification difficult with standard fixed-vector inputs. We propose a transformer-based model that represents each The Cancer Genome Atlas (TCGA) sample as a sequence of mutation tokens combining gene, genomic-position, substitution, sequence-context, variant-type, Variant Allele Fraction (VAF), read-depth, hotspot, and consequence features. Categorical fields were encoded by stable hashing, and numerical fields were normalized using only the training split of each fold. After filtering tumor types with at least 180 samples, 8,910 samples from 20 classes were evaluated. In stratified eight-fold cross-validation, the model achieved 64.4% accuracy, 65.4% weighted accuracy, 95.8% one-vs-rest ROC-AUC, 86.7% Top-3 accuracy, and 93.1% Top-5 accuracy. These results support sequence-based mutation modeling for ranked tumor-type prediction.

Null