Generative Artificial Intelligence Vs. Transformer-Based Framework for Classification of Cardiovascular Disease Patients using Transcriptomic Gene Data
Year: 2026
Type: conference-paper
Abstract: The classification of cardiovascular diseases using gene expression data is challenging due to high dimensionality, noise, and limited sample sizes. Traditional machine learning and deep learning models often struggle to generalize effectively in such settings, motivating the need for generative and energy-based approaches. We propose a Bio-GenAI framework that leverages generative artificial intelligence with denoising score matching to learn the underlying distribution of gene expression data. Noise is added to real samples, and the model learns to progressively restore them toward the real data manifold. An energy-based scoring mechanism is then used to classify case and control subjects. The proposed Bio-GenAI framework demonstrates improved classification performance compared with transformerbased models. The framework is evaluated on Hypertrophic Cardiomyopathy (HCM) and Acute Myocardial Infarction (AMI) datasets using K-fold validation. The proposed Bio-GenAI model achieved $\mathbf{\sim 9 8 \%}$ accuracy with an AUC of $\mathbf{0. 9 9}$ in seen paradigm. In Unseen paradigm, $\sim 94 \%$ accuracy with an AUC of 0.99, outperforming Transformers by 3.14% in HCM data and 5.2% in AMI data. All the models meet the regulatory requirements having $(\lt)10 \%$ difference between seen and unseen paradigm. The BioGenAI framework demonstrates robust performance and generalization, highlighting its potential for reliable gene-based cardiovascular disease classification.
Authors: Vandana Kumari, Alok Katiyar, Jasjit S. Suri
Institution: Galgotias University
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7167488486