Deep Learning based Classification of Copy Number Variations in Cervical Carcinoma using Array Comparative Genomic Hybridization Data

Deep learning's power model is evaluated for classifying genomic regions as cancerous or non-cancerous depends on copy number variation (CNV) patterns in cervical cancer, using high-resolution array comparative genomic hybridization (aCGH) data. The principal aim is to create a predictive model that identify segmental genomic alterations linked to cervical carcinogenesis. Utilize log2 ratio data from eight cervical cancer cell lines, the findings pursue to show the attainability of using artificial intelligence for genomic feature classification. A dataset consisting of 53,857 genomic probe measurements across eight cervical cancer cell lines is found from a GEO series matrix file. The log₂ intensity ratios representing CNVs were preprocessed and used as features. Synthetic binary labels generated by applying a threshold of ±0.25 on the average log₂ values, labeling probes with significant deviation as cancerous. A feedforward neural network comprising two hidden layers and dropout regularization is trained on this dataset. The proposed system is assessed using accuracy, precision, recall, F1-score, and confusion matrix analysis on a test set. The neural network achieved a test accuracy of 99.45%, with a precision of 0.90 and recall of 0.84 for the cancerous patients. Training and validation accuracy curves find close alignment, indicating minimal overfitting and strong model generalization. The results support the deep learning in analyzing CGH derived CNV data for cervical cancer research. This framework is a foundation for AI assisted genomic diagnostics and future clinical translation.