Enhanced Genetic Disorder Detection using LLM

Our work demonstrates a novel approach to predicting genetic disorders through state-of-the-art DNA-BERT models. We consider five significant genetic disorders: Marfan Syndrome, Cystic Fibrosis, Alzheimer's, Diabetes, and Breast Cancer. Experimental results demonstrate excellent performance, with the model achieving a 94.3% overall accuracy, precision of 92.8%, and recall of 93.1% across all disorders. In specific disorder analysis, the model performed particularly well in the detection of Alzheimer's disease with 95.2% accuracy, 94.8% precision, and a 94.6% F1-score. Similar excellent results were achieved for Cystic Fibrosis (94.8% accuracy, 93.7% precision, 94.2% F1-score) and Marfan Syndrome (94.5% accuracy, 93.8% precision, 94.1% F1-score). The model also performed well in detecting Diabetes (93.7% accuracy, 92.9% precision, 93.1% F1-score) and Breast Cancer (93.3% accuracy, 92.1% precision, 92.7% F1-score).The use of k-mer tokenization with the DNA-BERT model proved to be very effective in predicting diseases related to these disorders. The testing revealed that the model is reliable, with a ±1.2% confidence level, showing that it performs optimally across different datasets. The results reveal an improvement in the prediction of genetic disorders by quite a margin, which proves to be very useful for preventive healthcare's early detection and risk factor assessment. The model's high accuracy in predicting various disorders shows that it has the potential to become a strong tool in personalized medicine, allowing healthcare workers to carry out targeted interventions and treatments more effectively.The results of our study demonstrate not only how well our approach works in accurately predicting genetic disorders, but also its high potential to provide a good foundation for applying this to a variety of genetic diseases. It's valuable work to the field of genetic diagnosis and opens up avenues for applying machine learning to genomic medicine.