Transformer-Based Hybrid Model for DICER1 Syndrome Classification

Traditional computational methods often struggle to capture DNA's complex sequential and hierarchical nature, highlighting the need for advanced machine-learning techniques to analyze genomic data more accurately. Recent advancements in biological analysis and computer learning are transforming cancer treatment. AI can now match cancers to specific patients by utilizing deep DNA and radiological studies. Using deep learning (DL) techniques to forecast gene mutations from standard clinical tests is becoming increasingly popular, as detecting DICER1 syndrome in numerous patients requires significant time and expense. Although applying DL to medicine presents challenges, it holds promise in aiding oncologists in their decision-making. In our research, we developed a hybrid model that combines a Vision Transformer (ViT) with Long Short-Term Memory (LSTM) networks to classify mutations in the DICER1 gene. Our focus was on differentiating between mutations. The sequence data, consisting of around 71,000 nucleotides, was obtained from the NCBI database. We designed the model architecture to take advantage of both approaches: ViT excels at capturing long-range dependencies in the sequence, while the LSTM is well-suited for modeling the sequential patterns in the mutations. Together, they offered a comprehensive solution to the problem. By analyzing DICER1 germline mutations in DNA data, our model could significantly improve the early detection of pleuropulmonary blastomas (PPB). The results were outstanding, achieving an accuracy of 99.9%, precision of 99.9%, recall of 99.9%, F1-Score of 99.9%, and ROC-AUC of 1.0 on the test set. These findings highlight the potential of this hybrid model in accurately identifying gene mutations, which could be highly useful in genetic research, reducing misdiagnosis risks, and personalized medicine.