Shedding Light on the Black Box: Explaining Deep Neural Network Prediction of Clinical Outcomes.
Ontology highlight
ABSTRACT:
SUBMITTER: Shao Y
PROVIDER: S-EPMC7983057 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Age-reading of fish otoliths (ear stones) is important for the sustainable management of fish resources. However, the procedure is challenging and requires experienced readers to carefully examine annual growth zones. In a recent study, convolutional neural networks (CNNs) have been demonstrated to perform reasonably well on automatically predicting fish age from otolith images. In the present study, we carefully investigate the prediction rule learned by such neural networks to provide insight into the features that identify certain fish age ranges. For this purpose, a recent technique for visualizing and analyzing the predictions of the neural networks was applied to different versions of the otolith images. The results indicate that supplementary knowledge about the internal structure improves the results for the youngest age groups, compared to using only the contour shape attribute of the otolith. However, the contour shape and size attributes are, in general, sufficient for older age groups. In addition, within specific age ranges we find that the network tends to focus on particular areas of the otoliths and that the most discriminating factors seem to be related to the central part and the outer edge of the otolith. Explaining age predictions from otolith images as done in this study will hopefully help build confidence in the potential of deep learning algorithms for automatic age prediction, as well as improve the quality of the age estimation.
| S-EPMC7304622 | biostudies-literature
Project description:BackgroundThe study of the effectiveness of multidisciplinary rehabilitation (MDR) and how the symptoms and needs of individuals with multiple sclerosis (MS) interplay with the diversity of MDR interventions is still a conundrum, often referred to as a black box.MethodsWe conducted a partial crossover randomized controlled trial with follow-ups at 1 (discharge), 6, and 12 months. Based on their rehabilitation goals, each patient was categorized into 1 of 5 main focus areas (MFAs) prior to admission: Resilience, Cognitive Function, Energy, Physical Function, and Personal Needs. The Functional Assessment of Multiple Sclerosis (FAMS) instrument scores were the primary outcome.ResultsMFA groups varied in age (P = .036), MS type (P = .002), Expanded Disability Status Scale score (P < .001), time since diagnosis (P = .002), and FAMS at baseline (P < .001), as well as in composition and quantity of MDR services. At discharge, significant FAMS improvements were found in all 5 MFA groups (FAMS change > 10.4, P < .05), but the affected subdimensions and persistence of improvements varied among MFA groups. At the 6-month follow-up, estimates of controlled differences in FAMS were 9.9 (P =.001), 5.6 (P = .196), 8.5 (P = .008), -1.4 (P = .548), and 17.9 (P = .012) for the Resilience, Cognitive Function, Energy, Physical Function, and Personal Needs groups, respectively.ConclusionsThis study demonstrated that inpatient MDR improves functioning and health-related quality of life in people with MS; the type, degree, and persistence of the benefits are associated with a patient's main focus area of rehabilitation, which signifies the importance of the goal-setting process in MDR.
| S-EPMC11333915 | biostudies-literature
Project description:A molecule's geometry, also known as conformation, is one of a molecule's most important properties, determining the reactions it participates in, the bonds it forms, and the interactions it has with other molecules. Conventional conformation generation methods minimize hand-designed molecular force field energy functions that are often not well correlated with the true energy function of a molecule observed in nature. They generate geometrically diverse sets of conformations, some of which are very similar to the lowest-energy conformations and others of which are very different. In this paper, we propose a conditional deep generative graph neural network that learns an energy function by directly learning to generate molecular conformations that are energetically favorable and more likely to be observed experimentally in data-driven manner. On three large-scale datasets containing small molecules, we show that our method generates a set of conformations that on average is far more likely to be close to the corresponding reference conformations than are those obtained from conventional force field methods. Our method maintains geometrical diversity by generating conformations that are not too similar to each other, and is also computationally faster. We also show that our method can be used to provide initial coordinates for conventional force field methods. On one of the evaluated datasets we show that this combination allows us to combine the best of both methods, yielding generated conformations that are on average close to reference conformations with some very similar to reference conformations.
| S-EPMC6938476 | biostudies-literature
Project description:Motivation:A majority of known genetic variants associated with human-inherited diseases lie in non-coding regions that lack adequate interpretation, making it indispensable to systematically discover functional sites at the whole genome level and precisely decipher their implications in a comprehensive manner. Although computational approaches have been complementing high-throughput biological experiments towards the annotation of the human genome, it still remains a big challenge to accurately annotate regulatory elements in the context of a specific cell type via automatic learning of the DNA sequence code from large-scale sequencing data. Indeed, the development of an accurate and interpretable model to learn the DNA sequence signature and further enable the identification of causative genetic variants has become essential in both genomic and genetic studies. Results:We proposed Deopen, a hybrid framework mainly based on a deep convolutional neural network, to automatically learn the regulatory code of DNA sequences and predict chromatin accessibility. In a series of comparison with existing methods, we show the superior performance of our model in not only the classification of accessible regions against background sequences sampled at random, but also the regression of DNase-seq signals. Besides, we further visualize the convolutional kernels and show the match of identified sequence signatures and known motifs. We finally demonstrate the sensitivity of our model in finding causative noncoding variants in the analysis of a breast cancer dataset. We expect to see wide applications of Deopen with either public or in-house chromatin accessibility data in the annotation of the human genome and the identification of non-coding variants associated with diseases. Availability and implementation:Deopen is freely available at https://github.com/kimmo1019/Deopen. Contact:ruijiang@tsinghua.edu.cn. Supplementary information:Supplementary data are available at Bioinformatics online.
| S-EPMC6192215 | biostudies-literature
Project description:AimsTo develop a tool that can annotate subcellular localization of human proteins.BackgroundWith the progression of high throughput human proteomics projects, an enormous amount of protein sequence data has been discovered in the recent past. All these raw sequence data require precise mapping and annotation for their respective biological role and functional attributes. The functional characteristics of protein molecules are highly dependent on the subcellular localization/compartment. Therefore, a fully automated and reliable protein subcellular localization prediction system would be very useful for current proteomic research.ObjectiveTo develop a machine learning-based predictive model that can annotate the subcellular localization of human proteins with high accuracy and precision.MethodsIn this study, we used the PSI-CD-HIT homology criterion and utilized the sequence-based features of protein sequences to develop a powerful subcellular localization predictive model. The dataset used to train the HumDLoc model was extracted from a reliable data source, Uniprot knowledge base, which helps the model to generalize on the unseen dataset.ResultsThe proposed model, HumDLoc, was compared with two of the most widely used techniques: CELLO and DeepLoc, and other machine learning-based tools. The result demonstrated promising predictive performance of HumDLoc model based on various machine learning parameters such as accuracy (≥97.00%), precision (≥0.86), recall (≥0.89), MCC score (≥0.86), ROC curve (0.98 square unit), and precision-recall curve (0.93 square unit).ConclusionIn conclusion, HumDLoc was able to outperform several alternative tools for correctly predicting subcellular localization of human proteins. The HumDLoc has been hosted as a web-based tool at https://bioserver.iiita.ac.in/HumDLoc/.
| S-EPMC7604748 | biostudies-literature
Project description:Deep neural networks have led to state-of-the-art results in many medical imaging tasks including Alzheimer's disease (AD) detection based on structural magnetic resonance imaging (MRI) data. However, the network decisions are often perceived as being highly non-transparent, making it difficult to apply these algorithms in clinical routine. In this study, we propose using layer-wise relevance propagation (LRP) to visualize convolutional neural network decisions for AD based on MRI data. Similarly to other visualization methods, LRP produces a heatmap in the input space indicating the importance/relevance of each voxel contributing to the final classification outcome. In contrast to susceptibility maps produced by guided backpropagation ("Which change in voxels would change the outcome most?"), the LRP method is able to directly highlight positive contributions to the network classification in the input space. In particular, we show that (1) the LRP method is very specific for individuals ("Why does this person have AD?") with high inter-patient variability, (2) there is very little relevance for AD in healthy controls and (3) areas that exhibit a lot of relevance correlate well with what is known from literature. To quantify the latter, we compute size-corrected metrics of the summed relevance per brain area, e.g., relevance density or relevance gain. Although these metrics produce very individual "fingerprints" of relevance patterns for AD patients, a lot of importance is put on areas in the temporal lobe including the hippocampus. After discussing several limitations such as sensitivity toward the underlying model and computation parameters, we conclude that LRP might have a high potential to assist clinicians in explaining neural network decisions for diagnosing AD (and potentially other diseases) based on structural MRI data.
| S-EPMC6685087 | biostudies-literature
Project description:We present EP-DNN, a protocol for predicting enhancers based on chromatin features, in different cell types. Specifically, we use a deep neural network (DNN)-based architecture to extract enhancer signatures in a representative human embryonic stem cell type (H1) and a differentiated lung cell type (IMR90). We train EP-DNN using p300 binding sites, as enhancers, and TSS and random non-DHS sites, as non-enhancers. We perform same-cell and cross-cell predictions to quantify the validation rate and compare against two state-of-the-art methods, DEEP-ENCODE and RFECS. We find that EP-DNN has superior accuracy with a validation rate of 91.6%, relative to 85.3% for DEEP-ENCODE and 85.5% for RFECS, for a given number of enhancer predictions and also scales better for a larger number of enhancer predictions. Moreover, our H1 ? IMR90 predictions turn out to be more accurate than IMR90 ? IMR90, potentially because H1 exhibits a richer signature set and our EP-DNN model is expressive enough to extract these subtleties. Our work shows how to leverage the full expressivity of deep learning models, using multiple hidden layers, while avoiding overfitting on the training data. We also lay the foundation for exploration of cross-cell enhancer predictions, potentially reducing the need for expensive experimentation.
| S-EPMC5144062 | biostudies-literature
Project description:BackgroundImmunotherapy is an emerging approach in cancer treatment that activates the host immune system to destroy cancer cells expressing unique peptide signatures (neoepitopes). Administrations of cancer-specific neoepitopes in the form of synthetic peptide vaccine have been proven effective in both mouse models and human patients. Because only a tiny fraction of cancer-specific neoepitopes actually elicits immune response, selection of potent, immunogenic neoepitopes remains a challenging step in cancer vaccine development. A basic approach for immunogenicity prediction is based on the premise that effective neoepitope should bind with the Major Histocompatibility Complex (MHC) with high affinity.ResultsIn this study, we developed MHCSeqNet, an open-source deep learning model, which not only outperforms state-of-the-art predictors on both MHC binding affinity and MHC ligand peptidome datasets but also exhibits promising generalization to unseen MHC class I alleles. MHCSeqNet employed neural network architectures developed for natural language processing to model amino acid sequence representations of MHC allele and epitope peptide as sentences with amino acids as individual words. This consideration allows MHCSeqNet to accept new MHC alleles as well as peptides of any length.ConclusionsThe improved performance and the flexibility offered by MHCSeqNet should make it a valuable tool for screening effective neoepitopes in cancer vaccine development.
| S-EPMC6540523 | biostudies-literature
Project description:Deep learning-based tools may annotate and interpret medical data more quickly, consistently, and accurately than medical doctors. However, as medical doctors are ultimately responsible for clinical decision-making, any deep learning-based prediction should be accompanied by an explanation that a human can understand. We present an approach called electrocardiogram gradient class activation map (ECGradCAM), which is used to generate attention maps and explain the reasoning behind deep learning-based decision-making in ECG analysis. Attention maps may be used in the clinic to aid diagnosis, discover new medical knowledge, and identify novel features and characteristics of medical tests. In this paper, we showcase how ECGradCAM attention maps can unmask how a novel deep learning model measures both amplitudes and intervals in 12-lead electrocardiograms, and we show an example of how attention maps may be used to develop novel ECG features.
| S-EPMC8154909 | biostudies-literature
Project description:Residue solvent accessibility is closely related to the spatial arrangement and packing of residues. Predicting the solvent accessibility of a protein is an important step to understand its structure and function. In this work, we present a deep learning method to predict residue solvent accessibility, which is based on a stacked deep bidirectional recurrent neural network applied to sequence profiles. To capture more long-range sequence information, a merging operator was proposed when bidirectional information from hidden nodes was merged for outputs. Three types of merging operators were used in our improved model, with a long short-term memory network performing as a hidden computing node. The trained database was constructed from 7361 proteins extracted from the PISCES server using a cut-off of 25% sequence identity. Sequence-derived features including position-specific scoring matrix, physical properties, physicochemical characteristics, conservation score and protein coding were used to represent a residue. Using this method, predictive values of continuous relative solvent-accessible area were obtained, and then, these values were transformed into binary states with predefined thresholds. Our experimental results showed that our deep learning method improved prediction quality relative to current methods, with mean absolute error and Pearson's correlation coefficient values of 8.8% and 74.8%, respectively, on the CB502 dataset and 8.2% and 78%, respectively, on the Manesh215 dataset.
| S-EPMC6023031 | biostudies-literature