Table 1 Comparative analysis of existing and proposed methods.
From: Hybrid deep learning framework for accurate classification of high dimensional genomic data
Study | Approach | Data Type Used | Strengths | Limitations | Special Features |
|---|---|---|---|---|---|
El-Nabawy et al.13 | Feature-fusion for breast cancer classification | Clinical, genomic, histopathology data | Combines different data sources well | Suffers from data imbalance issues | Feature fusion |
Lu et al.14 | BrcaSeg – DL linking tissue images and genomics | Histopathology and genomic data | Maps image and genetic patterns | Needs broader validation studies | Tissue-genomic mapping |
Wei et al.15 | DeepTIS – two-stage DL for TIS prediction | Genomic sequences | Good at identifying initiation sites | Weak when sequences differ a lot | Two-stage design |
Huang et al.16 | ML applications for genomics-based therapy | Genomics for therapeutic tasks | Good coverage of therapy predictions | Interpretation challenges remain | Therapy-specific ML |
Ye et al.17 | Image-based DL for pan-cancer classification | Pan-cancer genomic data | High cancer detection rates | Hard to balance across cancer types | Image-genomics fusion |
Ahemad et al18. | ML methods for COVID-19 detection from genomics | Human genomic COVID-19 data | Good for early disease detection | Limited generalization | COVID-19 focus |
Erfanian et al19. | DL in single-cell genomics | Single-cell RNA and transcriptomics | Captures fine cellular variation | Black-box nature of models | Single-cell DL |
Bazgir and Lu20 | REFINED-CNN for survival prediction | High-dimensional genomic features | Handles complex feature spaces | Predictions are less explainable | Focus on survival analysis |
Khodaei et al21. | LPC and ML for viral genomic classification | Viral genomic sequences | Accurate virus classification | Sensitive to noise in sequences | LPC feature extraction |
Wang et al22. | DNNGP – DL for plant multi-omics data | Multi-omics plant datasets | Higher accuracy for genomic traits | Preprocessing is complex | Multi-omics fusion |
Zhu et al.23 | GSRNet with multi-scale CNN and BiGRU | Genomic signals and regions | Captures complex multi-scale features | Needs more computation and careful training | Adversarial training |
Abu-Doleh and Al Fahoum24 | XGBoost-CNN for cell type classification | Single-cell RNA-seq images | Good classification accuracy, useful for biological patterns | Hard to interpret results in large datasets | Image-based gene expression |
Mohammed et al.25 | U-Net based DNA sequence classifier | DNA sequence data | Strong feature extraction from sequences | Validation across multiple datasets is limited | Encoder-decoder structure |
Barber and Oueslati26 | Pre-trained ResNet-50, GoogleNet, and custom CNN | Exon and intron classification | Boosts classification with deep feature learning | Resource-heavy models | Use of multiple pre-trained networks |
Nawaz et al.27 | Sequential pattern discovery in genomic data | Sequential genomic data | Better sequence pattern recognition | Difficult to generalize across species | Sequential modeling |
Abbas et al.28 | Adaptive federated learning for cancer detection | Multi-source cancer data | Preserves data privacy | Model syncing challenges in real-world setups | Healthcare 5.0 integration |
Mora-Poblete et al29. | DL for phenomic-genomic modeling | Eucalyptus tree genomic data | Strong prediction by linking traits | Only for plants, limited species coverage | Phenomic-genomic integration |
Feng et al.30 | AI Breeder for genomic crop prediction | Crop genomic data | Genotype-phenotype integration | Needs careful tuning for diversity | AI-based breeding tool |
Batra et al.31 | AI pipeline for lung cancer screening | Radiology, clinical, genomic data | Combines multiple data types well | Managing privacy between modalities is complex | Early screening AI system |
Sangeetha et al.32 | Multimodal fusion DL for lung cancer | Multimodal lung cancer datasets | Boosts prediction through fusion | Handling missing data remains difficult | Enhanced multimodal fusion |
Raja et al.38 | Attention-guided CNN for liver tumor prediction | Genomic features | Focus on key genomic markers | Needs high-quality labeled data | Attention mechanism used |
Lin et al.39 | DL for fish phenotyping | Fish genomic and phenotypic data | Accurate prediction of physical traits | Focused only on tiger pufferfish | Genomic-based phenotyping |
Wang et al.40 | Cropformer – interpretable DL for crops | Crop genomic data | Combines performance and interpretability | Needs powerful computing resources | Transformer-based structure |
Wu et al.41 | AutoGP – AI for maize breeding programs | Maize genomic data | Good breeding predictions | Needs very large datasets | AI platform for breeding |
Proposed Method | TabNet & CNN with AFR | High-dimensional genomic data | Strong feature selection, robust learning, accurate classification | May need tuning for very sparse data | Combines TabNet’s feature selection with CNN’s DL power |
