Comparative Evaluation of Transformer and Hybrid BiLSTM-Transformer-MLP Models for Genomic Variant Pathogenicity Prediction
International Journal of Sciences and Innovation Engineering · 24.04.2026 · 3 (4) · 2171-2179
DOI: 10.70849/ijsci03042642273 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Sciences and Innovation Engineering, online erschienen am 24.04.2026. Verfasst von Dr.Kalavathi S und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Accurate prediction of genomic variant pathogenicity is important for computational genomics and clinical interpretation. This paper presents a comparative evaluation of a Transformer-based baseline and a proposed Hybrid BiLSTM-Transformer-MLP model for classifying ClinVar variants as benign or pathogenic. Mutation-centered 101 bp genomic sequences were extracted from the hg38 reference genome, one-hot encoded, and augmented with engineered biological features including GC content, transition/transversion flag, and reference/alternate allele encoding. In the proposed architecture, a BiLSTM branch captures sequential dependencies, a Transformer branch models contextual relationships through self-attention, and an MLP branch processes numerical features. The fused representation is passed through dense layers with sigmoid activation for binary classification. Experimental results show that the hybrid model outperforms the Transformer baseline across all evaluation metrics, achieving 0.89 accuracy, 0.85 precision, 0.87 recall, 0.932 F1-score, and 0.93 ROC-AUC, compared with 0.72, 0.65, 0.55, 0.71, and 0.61 respectively for the baseline.
Zitieren
APA
S, D., B, J., & N, J. (2026). Comparative Evaluation of Transformer and Hybrid BiLSTM-Transformer-MLP Models for Genomic Variant Pathogenicity Prediction. International Journal of Sciences and Innovation Engineering, 3(4), 2171-2179. https://doi.org/10.70849/ijsci03042642273
Vancouver
S D, B J, N J. Comparative Evaluation of Transformer and Hybrid BiLSTM-Transformer-MLP Models for Genomic Variant Pathogenicity Prediction. International Journal of Sciences and Innovation Engineering. 2026;3(4):2171-2179. doi:10.70849/ijsci03042642273
BibTeX
@article{S2026,
author = {S, Dr.Kalavathi and B, Janani and N, Jasmine},
title = {Comparative Evaluation of Transformer and Hybrid BiLSTM-Transformer-MLP Models for Genomic Variant Pathogenicity Prediction},
journal = {International Journal of Sciences and Innovation Engineering},
year = {2026},
volume = {3},
number = {4},
pages = {2171--2179},
doi = {10.70849/ijsci03042642273}
}
EUVIMED