Identifying potential small molecule-miRNA associations via Robust PCA based on γ-norm regularization.

Dysregulation of microRNAs (miRNAs) is closely associated with refractory human diseases, and the identification of potential associations between small molecule (SM) drugs and miRNAs can provide valuable insights for clinical treatment. Existing computational techniques for inferring potential associations suffer from limitations in terms of accuracy and efficiency. To address these challenges, we devise a novel predictive model called RPCA$\Gamma $NR, in which we propose a new Robust principal component analysis (PCA) framework based on $\gamma $-norm and $l_{2,1}$-norm regularization and design an Augmented Lagrange Multiplier method to optimize it, thereby deriving the association scores. The Gaussian Interaction Profile Kernel Similarity is calculated to capture the similarity information of SMs and miRNAs in known associations. Through extensive evaluation, including Cross Validation Experiments, Independent Validation Experiment, Efficiency Analysis, Ablation Experiment, Matrix Sparsity Analysis, and Case Studies, RPCA$\Gamma $NR outperforms state-of-the-art models concerning accuracy, efficiency and robustness. In conclusion, RPCA$\Gamma $NR can significantly streamline the process of determining SM-miRNA associations, thus contributing to advancements in drug development and disease treatment.

Generative Adversarial Matrix Completion Network based on Multi-Source Data Fusion for miRNA-Disease Associations Prediction.

Numerous biological studies have shown that considering disease-associated micro RNAs (miRNAs) as potential biomarkers or therapeutic targets offers new avenues for the diagnosis of complex diseases. Computational methods have gradually been introduced to reveal disease-related miRNAs. Considering that previous models have not fused sufficiently diverse similarities, that their inappropriate fusion methods may lead to poor quality of the comprehensive similarity network and that their results are often limited by insufficiently known associations, we propose a computational model called Generative Adversarial Matrix Completion Network based on Multi-source Data Fusion (GAMCNMDF) for miRNA-disease association prediction. We create a diverse network connecting miRNAs and diseases, which is then represented using a matrix. The main task of GAMCNMDF is to complete the matrix and obtain the predicted results. The main innovations of GAMCNMDF are reflected in two aspects: GAMCNMDF integrates diverse data sources and employs a nonlinear fusion approach to update the similarity networks of miRNAs and diseases. Also, some additional information is provided to GAMCNMDF in the form of a 'hint' so that GAMCNMDF can work successfully even when complete data are not available. Compared with other methods, the outcomes of 10-fold cross-validation on two distinct databases validate the superior performance of GAMCNMDF with statistically significant results. It is worth mentioning that we apply GAMCNMDF in the identification of underlying small molecule-related miRNAs, yielding outstanding performance results in this specific domain. In addition, two case studies about two important neoplasms show that GAMCNMDF is a promising prediction method.

Predicting potential small molecule-miRNA associations utilizing truncated schatten p-norm.

MicroRNAs (miRNAs) have significant implications in diverse human diseases and have proven to be effectively targeted by small molecules (SMs) for therapeutic interventions. However, current SM-miRNA association prediction models do not adequately capture SM/miRNA similarity. Matrix completion is an effective method for association prediction, but existing models use nuclear norm instead of rank function, which has some drawbacks. Therefore, we proposed a new approach for predicting SM-miRNA associations by utilizing the truncated schatten p-norm (TSPN). First, the SM/miRNA similarity was preprocessed by incorporating the Gaussian interaction profile kernel similarity method. This identified more SM/miRNA similarities and significantly improved the SM-miRNA prediction accuracy. Next, we constructed a heterogeneous SM-miRNA network by combining biological information from three matrices and represented the network with its adjacency matrix. Finally, we constructed the prediction model by minimizing the truncated schatten p-norm of this adjacency matrix and we developed an efficient iterative algorithmic framework to solve the model. In this framework, we also used a weighted singular value shrinkage algorithm to avoid the problem of excessive singular value shrinkage. The truncated schatten p-norm approximates the rank function more closely than the nuclear norm, so the predictions are more accurate. We performed four different cross-validation experiments on two separate datasets, and TSPN outperformed various most advanced methods. In addition, public literature confirms a large number of predictive associations of TSPN in four case studies. Therefore, TSPN is a reliable model for SM-miRNA association prediction.

CDK9 inhibition inhibits multiple oncogenic transcriptional and epigenetic pathways in prostate cancer.

BACKGROUND: Cyclin-dependent kinase 9 (CDK9) stimulates oncogenic transcriptional pathways in cancer and CDK9 inhibitors have emerged as promising therapeutic candidates. METHODS: The activity of an orally bioavailable CDK9 inhibitor, CDKI-73, was evaluated in prostate cancer cell lines, a xenograft mouse model, and patient-derived tumor explants and organoids. Expression of CDK9 was evaluated in clinical specimens by mining public datasets and immunohistochemistry. Effects of CDKI-73 on prostate cancer cells were determined by cell-based assays, molecular profiling and transcriptomic/epigenomic approaches. RESULTS: CDKI-73 inhibited proliferation and enhanced cell death in diverse in vitro and in vivo models of androgen receptor (AR)-driven and AR-independent models. Mechanistically, CDKI-73-mediated inhibition of RNA polymerase II serine 2 phosphorylation resulted in reduced expression of BCL-2 anti-apoptotic factors and transcriptional defects. Transcriptomic and epigenomic approaches revealed that CDKI-73 suppressed signaling pathways regulated by AR, MYC, and BRD4, key drivers of dysregulated transcription in prostate cancer, and reprogrammed cancer-associated super-enhancers. These latter findings prompted the evaluation of CDKI-73 with the BRD4 inhibitor AZD5153, a combination that was synergistic in patient-derived organoids and in vivo. CONCLUSION: Our work demonstrates that CDK9 inhibition disrupts multiple oncogenic pathways and positions CDKI-73 as a promising therapeutic agent for prostate cancer, particularly aggressive, therapy-resistant subtypes.

Radiotherapy with targeted and immunotherapy improved overall survival and progression-free survival for hepatocellular carcinoma with portal vein tumor thrombosis.

BACKGROUND: The efficacy of radiotherapy (RT) combined with targeted therapy and immunotherapy in treating hepatocellular carcinoma (HCC) and portal vein tumor thrombosis (PVTT) is still unclear. This study investigated the efficacy and safety of RT combined with targeted therapy and immunotherapy in HCC with PVTT. MATERIALS AND METHODS: Seventy-two patients with HCC with PVTT treated with tyrosine kinase inhibitor (TKI) plus programmed cell death protein-1 (PD-1) inhibitor with or without RT from December 2019 to December 2023 were included. After propensity score matching (PSM) for adjusting baseline differences, 32 pairs were identified in RT + TKI + PD-1 group (n = 32) and TKI + PD-1 group (n = 32). Primary endpoints were overall survival (OS) and progression-free survival (PFS). Secondary endpoints included objective response rate (ORR), disease control rate (DCR), and treatment-related adverse events (TRAEs). RESULTS: Median OS (mOS) in RT + TKI + PD-1 group was significantly longer than TKI + PD-1 group (15.6 vs. 8.2 months, P = .008). Median PFS (mPFS) in RT + TKI + PD-1 group was dramatically longer than TKI + PD-1 group (8.1 vs. 5.2 months, P = .011). Patients in TKI + PD-1 + RT group showed favorable ORR and DCR compared with TKI + PD-1 group (78.1% vs. 56.3%, P = .055; 93.8% vs. 81.3%, P = .128). Subgroup analysis demonstrated a remarkable OS and PFS benefit with TKI + PD-1 + RT for patients with main PVTT (type III/IV) and those of Child-Pugh class A. Multivariate analysis confirmed RT + TKI + PD-1 as an independent prognostic factor for longer OS (HR 0.391, P = .024) and longer PFS (HR 0.487, P = .013), with no mortality or severe TRAEs. CONCLUSION: RT combined with TKI and PD-1 inhibitor could significantly improve mOS and mPFS without inducing severe TRAEs or mortality.

Multirelational Hypergraph Representation Learning for Predicting circRNA-miRNA Associations.

One of the principal functions of circular RNA (circRNA) is to participate in gene regulation by sponging microRNAs (miRNAs). Using accumulated circRNA-miRNA associations (CMAs) to construct computational models for predicting potential associations provides a crucial tool for accelerating the validation of reliable associations through traditional experiments. Nevertheless, the current prediction models are constrained in their capacity to represent the higher-order relationships of CMAs and thus require further enhancement in terms of their predictive efficacy. In order to address this issue, we propose a new model based on multirelational hypergraph representation learning (MRHRL). This model employs hypergraphs to capture various higher-order relationships among RNAs and aggregates complementary information through a view attention mechanism. Furthermore, MRHRL introduces a hyperedge-level reconstruction task, jointly optimizing the prediction and reconstruction tasks within a unified framework to uncover potential information, thereby enhancing the model's predictive and generalization capabilities. Experiments conducted on three real-world data sets demonstrate that MRHRL achieves satisfactory results in CMAs prediction, significantly outperforming existing prediction models.

TIWMFLP: Two-Tier Interactive Weighted Matrix Factorization and Label Propagation Based on Similarity Matrix Fusion for Drug-Disease Association Prediction.

Accurately identifying new therapeutic uses for drugs is crucial for advancing pharmaceutical research and development. Matrix factorization is often used in association prediction due to its simplicity and high interpretability. However, existing matrix factorization models do not enable real-time interaction between molecular feature matrices and similarity matrices, nor do they consider the geometric structure of the matrices. Additionally, efficiently integrating multisource data remains a significant challenge. To address these issues, we propose a two-tier interactive weighted matrix factorization and label propagation model based on similarity matrix fusion (TIWMFLP) to assist in personalized treatment. First, we calculate the Gaussian and Laplace kernel similarities for drugs and diseases using known drug-disease associations. We then introduce a new multisource similarity fusion method, called similarity matrix fusion (SMF), to integrate these drug/disease similarities. SMF not only considers the different contributions represented by each neighbor but also incorporates drug-disease association information to enhance the contextual topological relationships and potential features of each drug/disease node in the network. Second, we innovatively developed a two-tier interactive weighted matrix factorization (TIWMF) method to process three biological networks. This method realizes for the first time the real-time interaction between the drug/disease feature matrix and its similarity matrix, allowing for a better capture of the complex relationships between drugs and diseases. Additionally, the weighted matrix of the drug/disease similarity matrix is introduced to preserve the underlying structure of the similarity matrix. Finally, the label propagation algorithm makes predictions based on the three updated biological networks. Experimental outcomes reveal that TIWMFLP consistently surpasses state-of-the-art models on four drug-disease data sets, two small molecule-miRNA data sets, and one miRNA-disease data set.

Connectivity Network Feature Sharing in Single-Cell RNA Sequencing Data Identifies Rare Cells.

Single-cell RNA sequencing is a valuable technique for identifying diverse cell subtypes. A key challenge in this process is that the detection of rare cells is often missed by conventional methods due to low abundance and subtle features of these cells. To overcome this, we developed SCLCNF (Local Connectivity Network Feature Sharing in Single-Cell RNA sequencing), a novel approach that identifies rare cells by analyzing features uniquely expressed in these cells. SCLCNF creates a cellular connectivity network, considering how each cell relates to its neighbors. This network helps to pinpoint coexpression patterns unique to rare cells, utilizing a rarity score to confirm their presence. Our method performs better in detecting rare cells than existing techniques, offering enhanced robustness. It has proven to be effective in human gastrula data sets for accurately pinpointing rare cells, and in sepsis data sets where it uncovers previously unidentified rare cell populations.

Targeting CDK2 to combat drug resistance in cancer therapy.

Drug resistance remains a major obstacle in cancer treatment, leading to treatment failures and high mortality rates. Despite advancements in therapies, overcoming resistance requires a deeper understanding of its mechanisms. This review highlights CDK2's pivotal role in both intrinsic and acquired resistance, and its potential as a therapeutic target. Cyclin E upregulation, which partners with CDK2, is linked to poor prognosis and resistance across various cancers. Specifically, amplifications of CCNE1/CCNE2 are associated with resistance to targeted therapies, immunotherapy, endocrine therapies and chemo/radiotherapy. Given CDK2's involvement in resistance mechanisms, investigating its role presents promising opportunities for developing novel strategies to combat resistance and improve treatment outcomes.

[Box: see text].

Markov state models elucidate the stability of DNA influenced by the chiral 5S-Tg base.

The static and dynamic structures of DNA duplexes affected by 5S-Tg (Tg, Thymine glycol) epimers were studied using MD simulations and Markov State Models (MSMs) analysis. The results show that the 5S,6S-Tg base caused little perturbation to the helix, and the base-flipping barrier was determined to be 4.4 kcal mol-1 through the use of enhanced sampling meta-eABF calculations, comparable to 5.4 kcal mol-1 of the corresponding thymine flipping. Two conformations with the different hydrogen bond structures between 5S,6R-Tg and A19 were identified in several independent MD trajectories. The 5S,6R-Tg:O6HO6•••N1:A19 hydrogen bond is present in the high-energy conformation displaying a clear helical distortion, and near barrier-free Tg base flipping. The low-energy conformation always maintains Watson-Crick base pairing between 5S,6R-Tg and A19, and 5S-Tg base flipping is accompanied by a small barrier of ca. 2.0 KBT (T = 298 K). The same conformations are observed in the MSMs analysis. Moreover, the transition path and metastable structures of the damaged base flipping are for the first time verified through MSMs analysis. The data clearly show that the epimers have completely different influence on the stability of the DNA duplex, thus implying different enzymatic mechanisms for DNA repair.

VGAEDTI: drug-target interaction prediction based on variational inference and graph autoencoder.

MOTIVATION: Accurate identification of Drug-Target Interactions (DTIs) plays a crucial role in many stages of drug development and drug repurposing. (i) Traditional methods do not consider the use of multi-source data and do not consider the complex relationship between data sources. (ii) How to better mine the hidden features of drug and target space from high-dimensional data, and better solve the accuracy and robustness of the model. RESULTS: To solve the above problems, a novel prediction model named VGAEDTI is proposed in this paper. We constructed a heterogeneous network with multiple sources of information using multiple types of drug and target dataIn order to obtain deeper features of drugs and targets, we use two different autoencoders. One is variational graph autoencoder (VGAE) which is used to infer feature representations from drug and target spaces. The second is graph autoencoder (GAE) propagating labels between known DTIs. Experimental results on two public datasets show that the prediction accuracy of VGAEDTI is better than that of six DTIs prediction methods. These results indicate that model can predict new DTIs and provide an effective tool for accelerating drug development and repurposing.

2-Anilino-4-(1-methyl-1H-pyrazol-4-yl)pyrimidine-derived CDK2 inhibitors as anticancer agents: Design, synthesis & evaluation.

Deregulation of cyclin-dependent kinase 2 (CDK2) and its activating partners, cyclins A and E, is associated with the pathogenesis of a myriad of human cancers and with resistance to anticancer drugs including CDK4/6 inhibitors. Thus, CDK2 has become an attractive target for the development of new anticancer therapies and for the amelioration of the resistance to CDK4/6 inhibitors. Bioisosteric replacement of the thiazole moiety of CDKI-73, a clinically trialled CDK inhibitor, by a pyrazole group afforded 9 and 19 that displayed potent CDK2-cyclin E inhibition (Ki = 0.023 and 0.001 µM, respectively) with submicromolar antiproliferative activity against a panel of cancer cell lines (GI50 = 0.025-0.780 µM). Mechanistic studies on 19 with HCT-116 colorectal cancer cells revealed that the compound reduced the phosphorylation of retinoblastoma at Ser807/811, arrested the cells at the G2/M phase, and induced apoptosis. These results highlight the potential of the 2-anilino-4-(1-methyl-1H-pyrazol-4-yl)pyrimidine series in developing potent and selective CDK2 inhibitors to combat cancer.

DeepFusion: A deep learning based multi-scale feature fusion method for predicting drug-target interactions.

Predicting drug-target interactions (DTIs) is essential for both drug discovery and drug repositioning. Recently, deep learning methods have achieved relatively significant performance in predicting DTIs. Generally, it needs a large amount of approved data of DTIs to train the model, which is actually tedious to obtain. In this work, we propose DeepFusion, a deep learning based multi-scale feature fusion method for predicting DTIs. To be specific, we generate global structural similarity feature based on similarity theory, convolutional neural network and generate local chemical sub-structure semantic feature using transformer network respectively for both drug and protein. Data experiments are conducted on four sub-datasets of BIOSNAP, which are 100%, 70%, 50% and 30% of BIOSNAP dataset. Particularly, using 70% sub-dataset, DeepFusion achieves ROC-AUC and PR-AUC by 0.877 and 0.888, which is close to the performance of some baseline methods trained by the whole dataset. In case study, DeepFusion achieves promising prediction results on predicting potential DTIs in case study.

MSGNN-DTA: Multi-Scale Topological Feature Fusion Based on Graph Neural Networks for Drug-Target Binding Affinity Prediction.

The accurate prediction of drug-target binding affinity (DTA) is an essential step in drug discovery and drug repositioning. Although deep learning methods have been widely adopted for DTA prediction, the complexity of extracting drug and target protein features hampers the accuracy of these predictions. In this study, we propose a novel model for DTA prediction named MSGNN-DTA, which leverages a fused multi-scale topological feature approach based on graph neural networks (GNNs). To address the challenge of accurately extracting drug and target protein features, we introduce a gated skip-connection mechanism during the feature learning process to fuse multi-scale topological features, resulting in information-rich representations of drugs and proteins. Our approach constructs drug atom graphs, motif graphs, and weighted protein graphs to fully extract topological information and provide a comprehensive understanding of underlying molecular interactions from multiple perspectives. Experimental results on two benchmark datasets demonstrate that MSGNN-DTA outperforms the state-of-the-art models in all evaluation metrics, showcasing the effectiveness of the proposed approach. Moreover, the study conducts a case study based on already FDA-approved drugs in the DrugBank dataset to highlight the potential of the MSGNN-DTA framework in identifying drug candidates for specific targets, which could accelerate the process of virtual screening and drug repositioning.

Discovery of N,4-Di(1H-pyrazol-4-yl)pyrimidin-2-amine-Derived CDK2 Inhibitors as Potential Anticancer Agents: Design, Synthesis, and Evaluation.

Cyclin-dependent kinase 2 (CDK2) has been garnering considerable interest as a target to develop new cancer treatments and to ameliorate resistance to CDK4/6 inhibitors. However, a selective CDK2 inhibitor has yet to be clinically approved. With the desire to discover novel, potent, and selective CDK2 inhibitors, the phenylsulfonamide moiety of our previous lead compound 1 was bioisosterically replaced with pyrazole derivatives, affording a novel series of N,4-di(1H-pyrazol-4-yl)pyrimidin-2-amines that exhibited potent CDK2 inhibitory activity. Among them, 15 was the most potent CDK2 inhibitor (Ki = 0.005 µM) with a degree of selectivity over other CDKs tested. Meanwhile, this compound displayed sub-micromolar antiproliferative activity against a panel of 13 cancer cell lines (GI50 = 0.127-0.560 µM). Mechanistic studies in ovarian cancer cells revealed that 15 reduced the phosphorylation of retinoblastoma at Thr821, arrested cells at the S and G2/M phases, and induced apoptosis. These results accentuate the potential of the N,4-di(1H-pyrazol-4-yl)pyrimidin-2-amine scaffold to be developed into potent and selective CDK2 inhibitors for the treatment of cancer.

CRANet: a comprehensive residual attention network for intracranial aneurysm image classification.

Rupture of intracranial aneurysm is the first cause of subarachnoid hemorrhage, second only to cerebral thrombosis and hypertensive cerebral hemorrhage, and the mortality rate is very high. MRI technology plays an irreplaceable role in the early detection and diagnosis of intracranial aneurysms and supports evaluating the size and structure of aneurysms. The increase in many aneurysm images, may be a massive workload for the doctors, which is likely to produce a wrong diagnosis. Therefore, we proposed a simple and effective comprehensive residual attention network (CRANet) to improve the accuracy of aneurysm detection, using a residual network to extract the features of an aneurysm. Many experiments have shown that the proposed CRANet model could detect aneurysms effectively. In addition, on the test set, the accuracy and recall rates reached 97.81% and 94%, which significantly improved the detection rate of aneurysms.

An overview of CDK3 in cancer: clinical significance and pharmacological implications.

Cyclin-dependent kinase 3 (CDK3) is a major player driving retinoblastoma (Rb) phosphorylation during the G0/G1 transition and in the early G1 phase of the cell cycle, preceding the effects of CDK4/cyclin D, CDK6/cyclin D, and CDK2/cyclin E. CDK3 can also directly regulate the activity of E2 factor (E2F) by skipping the role of Rb in late G1, potentially via the phosphorylation of the E2F1 partner DP1. Beyond the cell cycle, CDK3 interacts with various transcription factors involved in cell proliferation, differentiation, and transformation driven by the epidermal growth factor receptor (EGFR)/rat sarcoma virus (Ras) signaling pathway. The expression of CDK3 is extremely low in normal human tissue but upregulated in many cancers, implying a profound role in oncogenesis. Further evaluation of this role has been hampered by the lack of selective pharmacological inhibitors. Herein, we provide a comprehensive overview about the therapeutic potential of targeting CDK3 in cancer.

Repurposing existing therapeutics, its importance in oncology drug development: Kinases as a potential target.

Repurposing the large arsenal of existing non-cancer drugs is an attractive proposition to expand the clinical pipelines for cancer therapeutics. The earlier successes in repurposing resulted primarily from serendipitous findings, but more recently, drug or target-centric systematic identification of repurposing opportunities continues to rise. Kinases are one of the most sought-after anti-cancer drug targets over the last three decades. There are many non-cancer approved drugs that can inhibit kinases as "off-targets" as well as many existing kinase inhibitors that can target new additional kinases in cancer. Identifying cancer-associated kinase inhibitors through mining commercial drug databases or new kinase targets for existing inhibitors through comprehensive kinome profiling can offer more effective trial-ready options to rapidly advance drugs for clinical validation. In this review, we argue that drug repurposing is an important approach in modern drug development for cancer therapeutics. We have summarized the advantages of repurposing, the rationale behind this approach together with key barriers and opportunities in cancer drug development. We have also included examples of non-cancer drugs that inhibit kinases or are associated with kinase signalling as a basis for their anti-cancer action.

Molecular mechanisms of doxorubicin-induced cardiotoxicity: novel roles of sirtuin 1-mediated signaling pathways.

Doxorubicin (DOX) is an anthracycline chemotherapy drug used in the treatment of various types of cancer. However, short-term and long-term cardiotoxicity limits the clinical application of DOX. Currently, dexrazoxane is the only approved treatment by the United States Food and Drug Administration to prevent DOX-induced cardiotoxicity. However, a recent study found that pre-treatment with dexrazoxane could not fully improve myocardial toxicity of DOX. Therefore, further targeted cardioprotective prophylaxis and treatment strategies are an urgent requirement for cancer patients receiving DOX treatment to reduce the occurrence of cardiotoxicity. Accumulating evidence manifested that Sirtuin 1 (SIRT1) could play a crucially protective role in heart diseases. Recently, numerous studies have concentrated on the role of SIRT1 in DOX-induced cardiotoxicity, which might be related to the activity and deacetylation of SIRT1 downstream targets. Therefore, the aim of this review was to summarize the recent advances related to the protective effects, mechanisms, and deficiencies in clinical application of SIRT1 in DOX-induced cardiotoxicity. Also, the pharmaceutical preparations that activate SIRT1 and affect DOX-induced cardiotoxicity have been listed in this review.

SSGraphCPI: A Novel Model for Predicting Compound-Protein Interactions Based on Deep Learning.

Identifying compound-protein (drug-target, DTI) interactions (CPI) accurately is a key step in drug discovery. Including virtual screening and drug reuse, it can significantly reduce the time it takes to identify drug candidates and provide patients with timely and effective treatment. Recently, more and more researchers have developed CPI's deep learning model, including feature representation of a 2D molecular graph of a compound using a graph convolutional neural network, but this method loses much important information about the compound. In this paper, we propose a novel three-channel deep learning framework, named SSGraphCPI, for CPI prediction, which is composed of recurrent neural networks with an attentional mechanism and graph convolutional neural network. In our model, the characteristics of compounds are extracted from 1D SMILES string and 2D molecular graph. Using both the 1D SMILES string sequence and the 2D molecular graph can provide both sequential and structural features for CPI predictions. Additionally, we select the 1D CNN module to learn the hidden data patterns in the sequence to mine deeper information. Our model is much more suitable for collecting more effective information of compounds. Experimental results show that our method achieves significant performances with RMSE (Root Mean Square Error) = 2.24 and R2 (degree of linear fitting of the model) = 0.039 on the GPCR (G Protein-Coupled Receptors) dataset, and with RMSE = 2.64 and R2 = 0.018 on the GPCR dataset RMSE, which preforms better than some classical deep learning models, including RNN/GCNN-CNN, GCNNet and GATNet.