The research investigation evaluated machine learning technologies used for genomic analysis with a special focus on individual medical solutions

The gap

Evidence profile

Sourced from the recommendations and future work of the source papers, classified as general, drawn from work published between 2024 and 2026, spanning 3 journals. Those papers have been cited 11 times in total.

Supporting evidence — 3 representative gaps

• Exploring Machine Learning Applications for Genomic Data Analysis in Personalized Medicine (2026) · International Journal of Drug Delivery Technology · doi

  Conclusion The research investigation evaluated machine learning technologies used for genomic analysis with a special focus on individual medical solutions. Available research confirms how advanced deep learning approaches combined with tree-based models increase accuracy levels of predicting diseases while improving pharmaceutical advancement methods. CNNs surpassed traditional ML methods through their performance which yielded a 95.3% success rate in disease classification tasks. SHAP-based analyses feature determined the fundamental genetic variants responsible for disease predisposition which added to the models' interpretability capabilities. The application of Graph Neural Networks in drug discovery showed better results when predicting drug- target interactions because they generated MSE results at 0.012 which surpassed traditional docking algorithms. The results demonstrate AI methodology potential to research while enhancing quicken pharmaceutical targeted medical treatments. Reliable implementation of ML within genomic medicine remains conditional upon solving current data heterogeneity standards together with model understanding requirements and privacy privacy requirements.

  generalrecommendations
  Keywords: learning genomic medical based models predicting pharmaceutical surpassed traditional disease drug requirements privacy conclusion investigation
• Prediction of drug–disease associations based on reinforcement symmetric metric learning and graph convolution network (2024) · Frontiers in Pharmacology · cited 11× · doi

  system and GCN including GRGMF (Zhang et al., 2020), DRWBNCF (Meng et al., 2022), LAGCN (Yu et al., 2020b), DRHGCN (Cai et al., 2021) and CMLDR (Luo et al., 2021). These methods are detailed below. (cid:129) GRGMF establishes a generalized matrix factorization model that obtains the latent representation of each node by adaptively learning the neighborhood information of each node, and it introduces external similarity information to facilitate the prediction of potential links. (cid:129) DRWBNCF is a neural collaborative filtering method that proposes a new weighted bilinear graph convolution the known operation to integrate the information of drug–disease disease’s and association, neighborhood, and neighborhood interaction into a unified drug–disease representation to associations. infer novel potential drug’s (cid:129) LAGCN is a layer attention GCN that uses GCN to learn embeddings of drugs and diseases from the drug–disease heterogeneous network. The learned embeddings are then integrated by an attention mechanism to predict new associations. (cid:129) DRHGCN uses GCN to extract inter-domain and intra- domain feature information of drugs and diseases to find new drug indications based on different network topology information of drugs and diseases in different domains. (cid:129) CMLDR is a collaborative metric learning algorithm that predicts the association probability of drugs and diseases by applying metric learning. The latent vectors of drugs and diseases known related information of drugs and diseases and used to identify candidate drug–disease associations. learned based on the are For a fair comparison, we ran these competing methods with the optimal parameters suggested in the original papers on benchmark datasets. The complete evaluation of all methods was performed under 10-fold cross-validation. The specific experimental settings are described in Supplementary Material. Also, we conducted parameter analysis and selected the best parameters as for RSML-GCN in this work. recommended settings the 3.2 Parameter setting the further investigate influence Considering that hyperparameters could affect model performance, we of hyperparameters including that used in GCN, such as the latent vector dimension n, the marginal value strengths γ, and weight variables. The specific hyperparameter settings are given in Supplementary Material. According to the previous study (Yu et al., 2020a), we set the parameters for GCN with the embedding dimension k (cid:1) 64, number of layers L (cid:1) 3, initial learning rate lr1 (cid:1) 0.008, node discard rate β (cid:1) 0.6, regularize discard rate factor μ (cid:1) 6. Moreover, we have ξ (cid:1) 0.4, and penalty investigated the effect of the latent vector dimension n by varying its value from 30 to 400, and examined the influence of the marginal value strengths γ by varying its value from 0.01 to 100. The optimal parameters were determined by the grid search method, and detailed information

  generalrecommendations
  Keywords: information drug drugs diseases disease latent learning parameters value node neighborhood associations settings dimension rate
• Evolving computational paradigms for noncoding variant pathogenicity prediction (2026) · Frontiers in Molecular Biosciences · doi

  Looking ahead, several directions may help alleviate the challenges discussed above. First, future DL models should not focus solely on improving algorithms themselves but should also better incorporate the biological context and clinical knowledge. In other words, a closer collaboration between clinicians and algorithm researchers will be needed to define truly clinically meaningful questions, as well as appropriate model inputs and outputs. Second, multimodal learning is likely to become an important direction. Integrating genomic sequences with epigenomic information, three- dimensional genome organization, and single-cell data may provide a more comprehensive description of the biological context in which noncoding variants operate and may also help reduce the impact of data heterogeneity. Graph neural networks (GNNs) are another promising area worth further exploration. GNNs have already been applied to a range of disease-related inference and pathogenicity prediction tasks, including multitype variant classification, ncRNA/lncRNA–disease association prediction, and modeling of relationships between certain epigenetic sites and diseases (Ai et al., 2023; Farrokhi et al., 2026). However, these studies differ in both prediction targets and evaluation frameworks. For noncoding variants, one notable advantage of GNNs is their potential to extend local sequence information to higher-order regulatory networks and three-dimensional genome interactions. For example, GraphReg (Karbalayghareh et al., 2022) constructs regulatory graphs using chromatin conformation capture data and combines them with one-dimensional epigenomic signals or genomic sequences to predict gene expression, bringing enhancer–promoter interactions and distal regulatory information into a unified framework. More recently, multimodal graph models such as GNN-MAP (Yu et al., 2025) have also shown that integrating multiple layers of annotation for pathogenicity prediction is feasible, although these efforts have, so far, focused mainly on coding regions. Future graph-based models may further organize different layers of annotation into clearer and more interpretable multilayer graph structures. Combining GNNs with sequence encoders and epigenomic features may help better model distal regulatory effects and context-specific regulatory outputs. Given that graph convolutional networks (GCNs) (Hamed et al., 2025) have already been shown to effectively capture graph-structured features of DNA variant effects, their application to noncoding variant pathogenicity prediction remains highly promising. To address the shortage of labeled data, future studies may also place greater emphasis on few-shot learning and meta-learning, with the aim of extracting transferable and generalizable patterns from

  generalfuture work
  Keywords: graph prediction regulatory gnns help future models context learning epigenomic information dimensional noncoding networks pathogenicity

Questions about this gap