MOTIVATION: Accurate survival prediction is crucial for personalized cancer treatment but remains challenging for rare cancers due to limited data. Most deep learning models require large training datasets, which are unavailable for rare cancer types, creating a significant clini-cal bottleneck. RESULTS: We propose MoESurv, a zero-sample survival prediction framework that leverages a mix-ture-of-experts architecture to extract generalizable prognostic patterns from pan-cancer data. The model integrates shared experts, cancer-specific experts, and routing experts within an autoencoder to disentangle common and type-specific survival features. Evaluated on seven rare TCGA cancer types, MoESurv achieved state-of-the-art performance, improving the average C-index by 4 percentage points over the best baseline. Further-more, external validation across diverse populations and independent cohorts-including a Chinese glioma cohort (CGGA mRNAseq_693, C-index = 0.7433), a rare GBM IDH-mutant subtype (C-index = 0.8064), and the pan-cancer PCAWG cohort (C-index = 0.7090)-demonstrated that MoESurv possesses the most robust predictive per-formance, highlighting its generalizability. MoESurv also effectively stratified high- and low-risk patient groups and identified potential survival-associated genes, demonstrating both clinical utility and biological interpretability. AVAILABILITY: The code is freely available at https://github.com/HuaYC666/MoESurv and https://zenodo.org/records/20785500.

SUMMARY: Non-coding RNAs play important roles in transcriptional processes and participate in the regulation of various biological functions, in particular miRNAs and lncRNAs. Despite their importance for several biological functions, the existing signaling pathway databases do not include information on miRNA and lncRNA. Here, we redesigned a novel pathway database named NcPath by integrating and visualizing a total of 178 308 human experimentally validated miRNA-target interactions (MTIs), 32 282 experimentally verified lncRNA-target interactions (LTIs) and 4837 experimentally validated human ceRNA networks across 222 KEGG pathways (including 27 sub-categories). To expand the application potential of the redesigned NcPath database, we identified 556 798 reliable lncRNA-protein-coding genes (PCG) interaction pairs by integrating co-expression relations, ceRNA relations, co-TF-binding interactions, co-histone-modification interactions, cis-regulation relations and lncPro Tool predictions between lncRNAs and PCG. In addition, to determine the pathways in which miRNA/lncRNA targets are involved, we performed a KEGG enrichment analysis using a hypergeometric test. The NcPath database also provides information on MTIs/LTIs/ceRNA networks, PubMed IDs, gene annotations and the experimental verification method used. In summary, the NcPath database will serve as an important and continually updated platform that provides annotation and visualization of the pathways on which non-coding RNAs (miRNA and lncRNA) are involved, and provide support to multimodal non-coding RNAs enrichment analysis. The NcPath database is freely accessible at http://ncpath.pianlab.cn/. AVAILABILITY AND IMPLEMENTATION: NcPath database is freely available at http://ncpath.pianlab.cn/. The code and manual to use NcPath can be found at https://github.com/Marscolono/NcPath/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Protein lysine crotonylation (Kcr) is an important type of posttranslational modification that is associated with a wide range of biological processes. The identification of Kcr sites is critical to better understanding their functional mechanisms. However, the existing experimental techniques for detecting Kcr sites are cost-ineffective, to a great need for new computational methods to address this problem. We here describe Adapt-Kcr, an advanced deep learning model that utilizes adaptive embedding and is based on a convolutional neural network together with a bidirectional long short-term memory network and attention architecture. On the independent testing set, Adapt-Kcr outperformed the current state-of-the-art Kcr prediction model, with an improvement of 3.2% in accuracy and 1.9% in the area under the receiver operating characteristic curve. Compared to other Kcr models, Adapt-Kcr additionally had a more robust ability to distinguish between crotonylation and other lysine modifications. Another model (Adapt-ST) was trained to predict phosphorylation sites in SARS-CoV-2, and outperformed the equivalent state-of-the-art phosphorylation site prediction model. These results indicate that self-adaptive embedding features perform better than handcrafted features in capturing discriminative information; when used in attention architecture, this could be an effective way of identifying protein Kcr sites. Together, our Adapt framework (including learning embedding features and attention architecture) has a strong potential for prediction of other protein posttranslational modification sites.

BACKGROUND: Currently, large-scale gene expression profiling has been successfully applied to the discovery of functional connections among diseases, genetic perturbation, and drug action. To address the cost of an ever-expanding gene expression profile, a new, low-cost, high-throughput reduced representation expression profiling method called L1000 was proposed, with which one million profiles were produced. Although a set of ~ 1000 carefully chosen landmark genes that can capture ~ 80% of information from the whole genome has been identified for use in L1000, the robustness of using these landmark genes to infer target genes is not satisfactory. Therefore, more efficient computational methods are still needed to deep mine the influential genes in the genome. RESULTS: Here, we propose a computational framework based on deep learning to mine a subset of genes that can cover more genomic information. Specifically, an AutoEncoder framework is first constructed to learn the non-linear relationship between genes, and then DeepLIFT is applied to calculate gene importance scores. Using this data-driven approach, we have re-obtained a landmark gene set. The result shows that our landmark genes can predict target genes more accurately and robustly than that of L1000 based on two metrics [mean absolute error (MAE) and Pearson correlation coefficient (PCC)]. This reveals that the landmark genes detected by our method contain more genomic information. CONCLUSIONS: We believe that our proposed framework is very suitable for the analysis of biological big data to reveal the mysteries of life. Furthermore, the landmark genes inferred from this study can be used for the explosive amplification of gene expression profiles to facilitate research into functional connections.