Generative Artificial Intelligence vs. Transformer and Benchmarking Against Deep/Machine Learning: Classification and...

Area of Science:

• Cardiovascular Research
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Heart failure (HF) classification in women is complicated by sex-specific biological factors.
• Traditional models often fail to account for unique transcriptomic patterns in women.
• Acute myocardial infarction (AMI) gene expression data presents an opportunity for improved HF classification.

Purpose of the Study:

• To develop and validate a novel generative artificial intelligence (GenAI) framework for HF classification in women.
• To benchmark the GenAI model against transformer, deep learning (DL), and machine learning (ML) architectures.
• To leverage women's transcriptomic data for enhanced cardiovascular disease diagnosis.

Main Methods:

• Designed 26 models, including a novel wBio-GenAI, two transformers, 19 DL models (CNN, LSTM, xLSTM), and four ML models.
• Applied differential expression analysis (DEA) to identify differentially expressed genes (DEGs) from women's microarray data (GSE57345).
• Scientifically validated, benchmarked, and statistically tested the GenAI system's reliability.

Main Results:

• The wBio-GenAI model achieved 98.21% accuracy and an AUC of 0.99.
• wBio-GenAI outperformed transformer, DL, and ML models by significant margins (4.67%, 5.16%, and 15.07% respectively).
• The model met regulatory standards with a <10% difference between seen and unseen data.

Conclusions:

• The wBio-GenAI architecture effectively captures complex transcriptomic patterns in women.
• This AI model significantly improves heart failure classification accuracy in female patients.
• Advances are made in women-specific precision cardiovascular care through novel AI applications.