High-dimensional Biomedical Data Research Articles

Manifold Learning Uncovers Nonlinear Interactions Between the Adolescent Brain and Environment That Predict Emotional and Behavioral Problems

BackgroundTo progress adolescent mental health research beyond our present achievements—a complex account of brain and environmental risk factors without understanding neurobiological embedding in the environment—we need methods to uncover relationships between the developing brain and real-world environmental experiences. MethodsWe investigated associations between brain function, environments, and emotional and behavioral problems using participants from the Adolescent Brain Cognitive Development (ABCD) Study (n = 2401 female). We applied manifold learning, a promising technique for uncovering latent structure from high-dimensional biomedical data such as functional magnetic resonance imaging. Specifically, we developed exogenous PHATE (E-PHATE) to model brain-environment interactions. We used E-PHATE embeddings of participants’ brain activation during emotional and cognitive processing to predict individual differences in cognition and emotional and behavioral problems both cross-sectionally and longitudinally. ResultsE-PHATE embeddings of participants’ brain activation and environments at baseline showed moderate-to-large associations with total, externalizing, and internalizing problems at baseline, across several subcortical regions and large-scale cortical networks, compared with the zero-to-small effects achieved by voxel or PHATE methods. E-PHATE embeddings of the brain and environment at baseline were also related to emotional and behavioral problems 2 years later. These longitudinal predictions show a consistent moderate effect in the frontoparietal and attention networks. ConclusionsThe embedding of the adolescent brain in the environment yields enriched insight into emotional and behavioral problems. Using E-PHATE, we demonstrated how the harmonization of cutting-edge computational methods with longstanding developmental theories advances the detection and prediction of adolescent emotional and behavioral problems.

Manifold learning uncovers nonlinear interactions between the adolescent brain and environment that predict emotional and behavioral problems.

Background To progress adolescent mental health research beyond our present achievements - a complex account of brain and environmental risk factors without understanding neurobiological embedding in the environment - we need methods to unveil relationships between the developing brain and real-world environmental experiences. Methods We investigated associations among brain function, environments, and emotional and behavioral problems using participants from the Adolescent Brain and Cognitive Development Study (N=2,401 female). We applied manifold learning, a promising technique for uncovering latent structure from high-dimensional biomedical data like functional magnetic resonance imaging (fMRI). Specifically, we developed exogenous PHATE (E-PHATE) to model brain-environment interactions. We used E-PHATE embeddings of participants' brain activation during emotional and cognitive processing to predict individual differences in cognition and emotional and behavioral problems, both cross-sectionally and longitudinally. Results E-PHATE embeddings of participants' brain activation and environments at baseline show moderate-to-large associations with total, externalizing, and internalizing problems at baseline, across several subcortical regions and large-scale cortical networks, relative to the zero-to-small effects achieved by voxel or PHATE methods. E-PHATE embeddings of the brain and environment at baseline also relate to emotional and behavioral problems two years later. These longitudinal predictions show a consistent, moderate effect in the frontoparietal and attention networks. Conclusions Adolescent brain's embedding in the environment yields enriched insight into emotional and behavioral problems. Using E-PHATE, we demonstrate how the harmonization of cutting-edge computational methods with longstanding developmental theories advances detection and prediction of adolescent emotional and behavioral problems.

Feature selection based on dynamic crow search algorithm for high-dimensional data classification

High-dimensional biomedical data plays an important role in disease diagnosis and classification. To analyze the high-dimensional biomedical data, machine learning algorithms are widely used. However, a great amount of redundant, irrelevant or weak correlation features are prevalent in these datasets, which significantly deteriorate the capability of machine learning techniques. It is necessary to select the most informative features and discard the useless ones for enhancing the classification accuracy and reducing the dimensionality. To address this issue, an improved Crow Search Algorithm (CSA) which is known as the Dynamic CSA (DCSA) is proposed to find the optimal feature subset in high-dimensional classification tasks. In DCSA, three modifications are investigated. Firstly, we propose a dynamic bi-level awareness probability to guide the transition between global search and local search. Consequently, the balance between exploration and exploitation is improved. Secondly, levy flight with a unfixed step length control parameter is employed as the global search and the random search mechanism of the original CSA is abandoned. Thirdly, a dynamic flight length strategy is adopted to enhance the local search and speed up convergence. All in all, DCSA is developed as a wrapper feature selection model, where KNN classifier is applied as the feature subset evaluator. The performance of DCSA is measured on seven high-dimensional biomedical datasets. Experimental results show that DCSA outperforms other state-of-the-art methods. Specifically, DCSA achieved the highest average accuracy in six data. In terms of numbers of selected features, DCSA gained the smallest subset size in a seven data. And for SRBCT data, the prediction accuracy of DCSA is at 100%.

Covariance Matrix Estimation for High-Throughput Biomedical Data with Interconnected Communities

Estimating a covariance matrix is central to high-dimensional data analysis. Empirical analyses of high-dimensional biomedical data, including genomics, proteomics, microbiome, and neuroimaging, among others, consistently reveal strong modularity in the dependence patterns. In these analyses, intercorrelated high-dimensional biomedical features often form communities or modules that can be interconnected with others. While the interconnected community structure has been extensively studied in biomedical research (e.g., gene co-expression networks), its potential to assist in the estimation of covariance matrices remains largely unexplored. To address this gap, we propose a procedure that leverages the commonly observed interconnected community structure in high-dimensional biomedical data to estimate large covariance and precision matrices. We derive the uniformly minimum variance unbiased estimators for covariance and precision matrices in closed forms and provide theoretical results on their asymptotic properties. Our proposed method enhances the accuracy of covariance- and precision-matrix estimation and demonstrates superior performance compared to the competing methods in both simulations and real data analyses.

An Explainable AI System for the Diagnosis of High-Dimensional Biomedical Data

Typical state-of-the-art flow cytometry data samples typically consist of measures of 10 to 30 features of more than 100,000 cell “events”. Artificial intelligence (AI) systems are able to diagnose such data with almost the same accuracy as human experts. However, such systems face one central challenge: their decisions have far-reaching consequences for the health and lives of people. Therefore, the decisions of AI systems need to be understandable and justifiable by humans. In this work, we present a novel explainable AI (XAI) method called algorithmic population descriptions (ALPODS), which is able to classify (diagnose) cases based on subpopulations in high-dimensional data. ALPODS is able to explain its decisions in a form that is understandable to human experts. For the identified subpopulations, fuzzy reasoning rules expressed in the typical language of domain experts are generated. A visualization method based on these rules allows human experts to understand the reasoning used by the AI system. A comparison with a selection of state-of-the-art XAI systems shows that ALPODS operates efficiently on known benchmark data and on everyday routine case data.

An Efficient Binary Sand Cat Swarm Optimization for Feature Selection in High-Dimensional Biomedical Data.

Recent breakthroughs are making a significant contribution to big data in biomedicine which are anticipated to assist in disease diagnosis and patient care management. To obtain relevant information from this data, effective administration and analysis are required. One of the major challenges associated with biomedical data analysis is the so-called "curse of dimensionality". For this issue, a new version of Binary Sand Cat Swarm Optimization (called PILC-BSCSO), incorporating a pinhole-imaging-based learning strategy and crossover operator, is presented for selecting the most informative features. First, the crossover operator is used to strengthen the search capability of BSCSO. Second, the pinhole-imaging learning strategy is utilized to effectively increase exploration capacity while avoiding premature convergence. The Support Vector Machine (SVM) classifier with a linear kernel is used to assess classification accuracy. The experimental results show that the PILC-BSCSO algorithm beats 11 cutting-edge techniques in terms of classification accuracy and the number of selected features using three public medical datasets. Moreover, PILC-BSCSO achieves a classification accuracy of 100% for colon cancer, which is difficult to classify accurately, based on just 10 genes. A real Liver Hepatocellular Carcinoma (TCGA-HCC) data set was also used to further evaluate the effectiveness of the PILC-BSCSO approach. PILC-BSCSO identifies a subset of five marker genes, including prognostic biomarkers HMMR, CHST4, and COL15A1, that have excellent predictive potential for liver cancer using TCGA data.

Statistical analysis of high-dimensional biomedical data: a gentle introduction to analytical goals, common approaches and challenges

BackgroundIn high-dimensional data (HDD) settings, the number of variables associated with each observation is very large. Prominent examples of HDD in biomedical research include omics data with a large number of variables such as many measurements across the genome, proteome, or metabolome, as well as electronic health records data that have large numbers of variables recorded for each patient. The statistical analysis of such data requires knowledge and experience, sometimes of complex methods adapted to the respective research questions.MethodsAdvances in statistical methodology and machine learning methods offer new opportunities for innovative analyses of HDD, but at the same time require a deeper understanding of some fundamental statistical concepts. Topic group TG9 “High-dimensional data” of the STRATOS (STRengthening Analytical Thinking for Observational Studies) initiative provides guidance for the analysis of observational studies, addressing particular statistical challenges and opportunities for the analysis of studies involving HDD. In this overview, we discuss key aspects of HDD analysis to provide a gentle introduction for non-statisticians and for classically trained statisticians with little experience specific to HDD.ResultsThe paper is organized with respect to subtopics that are most relevant for the analysis of HDD, in particular initial data analysis, exploratory data analysis, multiple testing, and prediction. For each subtopic, main analytical goals in HDD settings are outlined. For each of these goals, basic explanations for some commonly used analysis methods are provided. Situations are identified where traditional statistical methods cannot, or should not, be used in the HDD setting, or where adequate analytic tools are still lacking. Many key references are provided.ConclusionsThis review aims to provide a solid statistical foundation for researchers, including statisticians and non-statisticians, who are new to research with HDD or simply want to better evaluate and understand the results of HDD analyses.

SIMON, a powerful AI software accelerating discovery of human immune memory responses to viruses

Abstract Influenza and SARS-CoV-2 are among the greatest viral threats to human health causing the 20th and 21st centuries’ most significant pandemics. These viruses have caused millions of deaths worldwide and continue to pose a significant risk to global health. To effectively fight against these viruses, it is essential to understand the complexity behind the generation of immune memory responses. The traditional analytical methods are often inadequate for capturing the complexity of the immune responses. To address this challenge, we have developed SIMON, a new AI software that can perform integrative multi-omics analysis to study human immune memory responses. With an easy-to-use graphical user interface, standardized pipelines, automated approach for machine learning testing over 180 algorithms, and other statistical analysis methods, SIMON helps to identify crucial patterns in biomedical data. We have demonstrated the accuracy, ease of use, and power of our software on various biomedical datasets, gaining valuable insights into the mechanisms underlying immunity to SARS-CoV-2 and influenza. This knowledge can be used to develop more effective vaccines and help researchers better understand the long-term impact of SARS-CoV-2 or influenza viruses on the immune system. By predicting the immune memory responses, SIMON has the potential to accelerate the development of vaccines that provide long-term protection against these pandemic viruses. Overall, SIMON is a powerful software platform for data mining that facilitates pattern recognition and knowledge extraction from high-dimensional biomedical data. It has the potential to greatly accelerate discovery in human immunology and biomedicine.

A Review on Nature-Inspired Algorithms for Cancer Disease Prediction and Classification

In the era of healthcare and its related research fields, the dimensionality problem of high-dimensional data is a massive challenge as it is crucial to identify significant genes while conducting research on diseases like cancer. As a result, studying new Machine Learning (ML) techniques for raw gene expression biomedical data is an important field of research. Disease detection, sample classification, and early disease prediction are all important analyses of high-dimensional biomedical data in the field of bioinformatics. Recently, machine-learning techniques have dramatically improved the analysis of high-dimension biomedical data sets. Nonetheless, researchers’ studies on biomedical data faced the challenge of vast dimensions, i.e., the vast features (genes) with a very low sample space. In this paper, two-dimensionality reduction methods, feature selection, and feature extraction are introduced with a systematic comparison of several dimension reduction techniques for the analysis of high-dimensional gene expression biomedical data. We presented a systematic review of some of the most popular nature-inspired algorithms and analyzed them. The paper is mainly focused on the original principles behind each of the algorithms and their applications for cancer classification and prediction from gene expression data. Lastly, the advantages and disadvantages of nature-inspired algorithms for biomedical data are evaluated. This review paper may guide researchers to choose the most effective algorithm for cancer classification and prediction for the satisfactory analysis of high-dimensional biomedical data.

CoIn: Correlation Induced Clustering for Cognition of High Dimensional Bioinformatics Data.

Analysis of high dimensional biomedical data such as microarray gene expression data and mass spectrometry images, is crucial to provide better medical services including cancer subtyping, protein homology detection,etc. Clustering is a fundamental cognitive task which aims to group unlabeled data into multiple clusters based on their intrinsic similarities. The K-means algorithm is one of the most widely used clustering heuristics that aims at grouping the data objects into meaningful clusters such that the sum of squared Euclidean distances within each cluster is minimized. Its conceptual simplicity and computational efficiency make it easy to be used for wide applications of different data types. However, all features of data in K-means are considered equally in relevance, which distorts the performance when clustering high-dimensional data such as microarray gene expression data, mass spectrometry images, where there exist many redundant variables and correlated variables. In this paper, we propose a new correlation induced clustering, CoIn, to capture complex correlations among high dimensional data and guarantee the correlation consistency within each cluster. We evaluate the proposed method on a high dimensional mass spectrometry dataset of liver cancer tumor to explore the metabolic differences on tissues and discover the intra-tumor heterogeneity (ITH). By comparing the results of baselines and ours, it has been found that our method produces more explainable and understandable results for clinical analysis, which demonstrates the proposed clustering paradigm has the potential with application to knowledge discovery in high dimensional bioinformatics data.

Robust statistical boosting with quantile-based adaptive loss functions.

We combine robust loss functions with statistical boosting algorithms in an adaptive way to perform variable selection and predictive modelling for potentially high-dimensional biomedical data. To achieve robustness against outliers in the outcome variable (vertical outliers), we consider different composite robust loss functions together with base-learners for linear regression. For composite loss functions, such as the Huber loss and the Bisquare loss, a threshold parameter has to be specified that controls the robustness. In the context of boosting algorithms, we propose an approach that adapts the threshold parameter of composite robust losses in each iteration to the current sizes of residuals, based on a fixed quantile level. We compared the performance of our approach to classical M-regression, boosting with standard loss functions or the lasso regarding prediction accuracy and variable selection in different simulated settings: the adaptive Huber and Bisquare losses led to a better performance when the outcome contained outliers or was affected by specific types of corruption. For non-corrupted data, our approach yielded a similar performance to boosting with the efficient L 2 loss or the lasso. Also in the analysis of skewed KRT19 protein expression data based on gene expression measurements from human cancer cell lines (NCI-60 cell line panel), boosting with the new adaptive loss functions performed favourably compared to standard loss functions or competing robust approaches regarding prediction accuracy and resulted in very sparse models.

Hybrid binary arithmetic optimization algorithm with simulated annealing for feature selection in high-dimensional biomedical data

Hybrid binary arithmetic optimization algorithm with simulated annealing for feature selection in high-dimensional biomedical data

Feature Selection in High Dimensional Biomedical Data Based on BF-SFLA.

High-dimensional biomedical data contained many irrelevant or weakly correlated features, which affected the efficiency of disease diagnosis. This manuscript presented a feature selection method for high-dimensional biomedical data based on the chemotaxis foraging-shuffled frog leaping algorithm (BF-SFLA). The performance of the BF-SFLA based feature selection method was further improved by introducing chemokine operation and balanced grouping strategies into the shuffled frog leaping algorithm, which maintained the balance between global optimization and local optimization and reduced the possibility of the algorithm falling into local optimization. To evaluate the proposed method’s effectiveness, we employed the K-NN (k-nearest Neighbor) and C4.5 decision tree classification algorithm with a comparative analysis. We compared our proposed approach with improved genetic algorithms, particle swarm optimization, and the basic shuffled frog leaping algorithm. Experimental results showed that the feature selection method based on BF-SFLA obtained a better feature subset, improved classification accuracy, and shortened classification time.

A Novel Feature Selection Method Based on MRMR and Enhanced Flower Pollination Algorithm for High Dimensional Biomedical Data

Background: The massive amount of biomedical data accumulated in the past decades can be utilized for diagnosing disease. Objective: However, the high dimensionality, small sample sizes, and irrelevant features of data often have a negative influence on the accuracy and speed of disease prediction. Some existing machine learning models cannot capture the patterns on these datasets accurately without utilizing feature selection. Methods: Filter and wrapper are two prevailing feature selection methods. The filter method is fast but has low prediction accuracy, while the latter can obtain high accuracy but has a formidable computation cost. Given the drawbacks of using filter or wrapper individually, a novel feature selection method, called MRMR-EFPATS, is proposed, which hybridizes filter method Minimum Redundancy Maximum Relevance (MRMR) and wrapper method based on an improved Flower Pollination Algorithm (FPA). First, MRMR is employed to rank and screen out some important features quickly. These features are further chosen for individual populations following the wrapper method for faster convergence and less computational time. Then, due to its efficiency and flexibility, FPA is adopted to further discover an optimal feature subset. Result: FPA still has some drawbacks, such as slow convergence rate, inadequacy in terms of searching new solutions, and tends to be trapped in local optima. In our work, an elite strategy is adopted to improve the convergence speed of the FPA. Tabu search and Adaptive Gaussian Mutation are employed to improve the search capability of FPA and escape from local optima. Here, the KNN classifier with the 5-fold-CV is utilized to evaluate the classification accuracy. Conclusion: Extensive experimental results on six public high dimensional biomedical datasets show that the proposed MRMR-EFPATS has achieved superior performance compared to other state-of-theart methods.

Determination of Survival of Gastric Cancer Patients With Distant Lymph Node Metastasis Using Prealbumin Level and Prothrombin Time: Contour Plots Based on Random Survival Forest Algorithm on High-Dimensionality Clinical and Laboratory Datasets

PurposeThis study aimed to identify prognostic factors for patients with distant lymph node-involved gastric cancer (GC) using a machine learning algorithm, a method that offers considerable advantages and new prospects for high-dimensional biomedical data exploration.Materials and MethodsThis study employed 79 features of clinical pathology, laboratory tests, and therapeutic details from 289 GC patients whose distant lymphadenopathy was presented as the first episode of recurrence or metastasis. Outcomes were measured as any-cause death events and survival months after distant lymph node metastasis. A prediction model was built based on possible outcome predictors using a random survival forest algorithm and confirmed by 5×5 nested cross-validation. The effects of single variables were interpreted using partial dependence plots. A contour plot was used to visually represent survival prediction based on 2 predictive features.ResultsThe median survival time of patients with GC with distant nodal metastasis was 9.2 months. The optimal model incorporated the prealbumin level and the prothrombin time (PT), and yielded a prediction error of 0.353. The inclusion of other variables resulted in poorer model performance. Patients with higher serum prealbumin levels or shorter PTs had a significantly better prognosis. The predicted one-year survival rate was stratified and illustrated as a contour plot based on the combined effect the prealbumin level and the PT.ConclusionsMachine learning is useful for identifying the important determinants of cancer survival using high-dimensional datasets. The prealbumin level and the PT on distant lymph node metastasis are the 2 most crucial factors in predicting the subsequent survival time of advanced GC.Trial RegistrationChiCTR Identifier: ChiCTR1800019978

Two-stage feature selection for classification of gene expression data based on an improved Salp Swarm Algorithm.

Microarray technology has developed rapidly in recent years, producing a large number of ultra-high dimensional gene expression data. However, due to the huge sample size and dimension proportion of gene expression data, it is very challenging work to screen important genes from gene expression data. For small samples of high-dimensional biomedical data, this paper proposes a two-stage feature selection framework combining Wrapper, embedding and filtering to avoid the curse of dimensionality. The proposed framework uses weighted gene co-expression network (WGCNA), random forest and minimal redundancy maximal relevance (mRMR) for first stage feature selection. In the second stage, a new gene selection method based on the improved binary Salp Swarm Algorithm is proposed, which combines machine learning methods to adaptively select feature subsets suitable for classification algorithms. Finally, the classification accuracy is evaluated using six methods: lightGBM, RF, SVM, XGBoost, MLP and KNN. To verify the performance of the framework and the effectiveness of the proposed algorithm, the number of genes selected and the classification accuracy was compared with the other five intelligent optimization algorithms. The results show that the proposed framework achieves an accuracy equal to or higher than other advanced intelligent algorithms on 10 datasets, and achieves an accuracy of over 97.6% on all 10 datasets. This shows that the method proposed in this paper can solve the feature selection problem related to high-dimensional data, and the proposed framework has no data set limitation, and it can be applied to other fields involving feature selection.

A scalable software solution for anonymizing high-dimensional biomedical data.

BackgroundData anonymization is an important building block for ensuring privacy and fosters the reuse of data. However, transforming the data in a way that preserves the privacy of subjects while maintaining a high degree of data quality is challenging and particularly difficult when processing complex datasets that contain a high number of attributes. In this article we present how we extended the open source software ARX to improve its support for high-dimensional, biomedical datasets.FindingsFor improving ARX's capability to find optimal transformations when processing high-dimensional data, we implement 2 novel search algorithms. The first is a greedy top-down approach and is oriented on a formally implemented bottom-up search. The second is based on a genetic algorithm. We evaluated the algorithms with different datasets, transformation methods, and privacy models. The novel algorithms mostly outperformed the previously implemented bottom-up search. In addition, we extended the GUI to provide a high degree of usability and performance when working with high-dimensional datasets.ConclusionWith our additions we have significantly enhanced ARX's ability to handle high-dimensional data in terms of processing performance as well as usability and thus can further facilitate data sharing.

Statistical Enrichment Analysis of Samples: A General-Purpose Tool to Annotate Metadata Neighborhoods of Biological Samples.

Unsupervised learning techniques, such as clustering and embedding, have been increasingly popular to cluster biomedical samples from high-dimensional biomedical data. Extracting clinical data or sample meta-data shared in common among biomedical samples of a given biological condition remains a major challenge. Here, we describe a powerful analytical method called Statistical Enrichment Analysis of Samples (SEAS) for interpreting clustered or embedded sample data from omics studies. The method derives its power by focusing on sample sets, i.e., groups of biological samples that were constructed for various purposes, e.g., manual curation of samples sharing specific characteristics or automated clusters generated by embedding sample omic profiles from multi-dimensional omics space. The samples in the sample set share common clinical measurements, which we refer to as “clinotypes,” such as age group, gender, treatment status, or survival days. We demonstrate how SEAS yields insights into biological data sets using glioblastoma (GBM) samples. Notably, when analyzing the combined The Cancer Genome Atlas (TCGA)—patient-derived xenograft (PDX) data, SEAS allows approximating the different clinical outcomes of radiotherapy-treated PDX samples, which has not been solved by other tools. The result shows that SEAS may support the clinical decision. The SEAS tool is publicly available as a freely available software package at https://aimed-lab.shinyapps.io/SEAS/.

Randomized boosting with multivariable base-learners for high-dimensional variable selection and prediction

BackgroundStatistical boosting is a computational approach to select and estimate interpretable prediction models for high-dimensional biomedical data, leading to implicit regularization and variable selection when combined with early stopping. Traditionally, the set of base-learners is fixed for all iterations and consists of simple regression learners including only one predictor variable at a time. Furthermore, the number of iterations is typically tuned by optimizing the predictive performance, leading to models which often include unnecessarily large numbers of noise variables.ResultsWe propose three consecutive extensions of classical component-wise gradient boosting. In the first extension, called Subspace Boosting (SubBoost), base-learners can consist of several variables, allowing for multivariable updates in a single iteration. To compensate for the larger flexibility, the ultimate selection of base-learners is based on information criteria leading to an automatic stopping of the algorithm. As the second extension, Random Subspace Boosting (RSubBoost) additionally includes a random preselection of base-learners in each iteration, enabling the scalability to high-dimensional data. In a third extension, called Adaptive Subspace Boosting (AdaSubBoost), an adaptive random preselection of base-learners is considered, focusing on base-learners which have proven to be predictive in previous iterations. Simulation results show that the multivariable updates in the three subspace algorithms are particularly beneficial in cases of high correlations among signal covariates. In several biomedical applications the proposed algorithms tend to yield sparser models than classical statistical boosting, while showing a very competitive predictive performance also compared to penalized regression approaches like the (relaxed) lasso and the elastic net.ConclusionsThe proposed randomized boosting approaches with multivariable base-learners are promising extensions of statistical boosting, particularly suited for highly-correlated and sparse high-dimensional settings. The incorporated selection of base-learners via information criteria induces automatic stopping of the algorithms, promoting sparser and more interpretable prediction models.

DeepFeature: feature selection in nonimage data using convolutional neural network.

Artificial intelligence methods offer exciting new capabilities for the discovery of biological mechanisms from raw data because they are able to detect vastly more complex patterns of association that cannot be captured by classical statistical tests. Among these methods, deep neural networks are currently among the most advanced approaches and, in particular, convolutional neural networks (CNNs) have been shown to perform excellently for a variety of difficult tasks. Despite that applications of this type of networks to high-dimensional omics data and, most importantly, meaningful interpretation of the results returned from such models in a biomedical context remains an open problem. Here we present, an approach applying a CNN to nonimage data for feature selection. Our pipeline, DeepFeature, can both successfully transform omics data into a form that is optimal for fitting a CNN model and can also return sets of the most important genes used internally for computing predictions. Within the framework, the Snowfall compression algorithm is introduced to enable more elements in the fixed pixel framework, and region accumulation and element decoder is developed to find elements or genes from the class activation maps. In comparative tests for cancer type prediction task, DeepFeature simultaneously achieved superior predictive performance and better ability to discover key pathways and biological processes meaningful for this context. Capabilities offered by the proposed framework can enable the effective use of powerful deep learning methods to facilitate the discovery of causal mechanisms in high-dimensional biomedical data.