Towards On-Device Dehydration Monitoring Using Machine Learning from Wearable Device's Data.

With the ongoing advances in sensor technology and miniaturization of electronic chips, more applications are researched and developed for wearable devices. Hydration monitoring is among the problems that have been recently researched. Athletes, battlefield soldiers, workers in extreme weather conditions, people with adipsia who have no sensation of thirst, and elderly people who lost their ability to talk are among the main target users for this application. In this paper, we address the use of machine learning for hydration monitoring using data from wearable sensors: accelerometer, magnetometer, gyroscope, galvanic skin response sensor, photoplethysmography sensor, temperature, and barometric pressure sensor. These data, together with new features constructed to reflect the activity level, were integrated with personal features to predict the last drinking time of a person and alert the user when it exceeds a certain threshold. The results of applying different models are compared for model selection for on-device deployment optimization. The extra trees model achieved the least error for predicting unseen data; random forest came next with less training time, then the deep neural network with a small model size, which is preferred for wearable devices with limited memory. Embedded on-device testing is still needed to emphasize the results and test for power consumption.

Interpreting patient-Specific risk prediction using contextual decomposition of BiLSTMs: application to children with asthma.

BACKGROUND: Predictive modeling with longitudinal electronic health record (EHR) data offers great promise for accelerating personalized medicine and better informs clinical decision-making. Recently, deep learning models have achieved state-of-the-art performance for many healthcare prediction tasks. However, deep models lack interpretability, which is integral to successful decision-making and can lead to better patient care. In this paper, we build upon the contextual decomposition (CD) method, an algorithm for producing importance scores from long short-term memory networks (LSTMs). We extend the method to bidirectional LSTMs (BiLSTMs) and use it in the context of predicting future clinical outcomes using patients' EHR historical visits. METHODS: We use a real EHR dataset comprising 11071 patients, to evaluate and compare CD interpretations from LSTM and BiLSTM models. First, we train LSTM and BiLSTM models for the task of predicting which pre-school children with respiratory system-related complications will have asthma at school-age. After that, we conduct quantitative and qualitative analysis to evaluate the CD interpretations produced by the contextual decomposition of the trained models. In addition, we develop an interactive visualization to demonstrate the utility of CD scores in explaining predicted outcomes. RESULTS: Our experimental evaluation demonstrate that whenever a clear visit-level pattern exists, the models learn that pattern and the contextual decomposition can appropriately attribute the prediction to the correct pattern. In addition, the results confirm that the CD scores agree to a large extent with the importance scores generated using logistic regression coefficients. Our main insight was that rather than interpreting the attribution of individual visits to the predicted outcome, we could instead attribute a model's prediction to a group of visits. CONCLUSION: We presented a quantitative and qualitative evidence that CD interpretations can explain patient-specific predictions using CD attributions of individual visits or a group of visits.

Assessment of de novo assemblers for draft genomes: a case study with fungal genomes.

BACKGROUND: Recently, large bio-projects dealing with the release of different genomes have transpired. Most of these projects use next-generation sequencing platforms. As a consequence, many de novo assembly tools have evolved to assemble the reads generated by these platforms. Each tool has its own inherent advantages and disadvantages, which make the selection of an appropriate tool a challenging task. RESULTS: We have evaluated the performance of frequently used de novo assemblers namely ABySS, IDBA-UD, Minia, SOAP, SPAdes, Sparse, and Velvet. These assemblers are assessed based on their output quality during the assembly process conducted over fungal data. We compared the performance of these assemblers by considering both computational as well as quality metrics. By analyzing these performance metrics, the assemblers are ranked and a procedure for choosing the candidate assembler is illustrated. CONCLUSIONS: In this study, we propose an assessment method for the selection of de novo assemblers by considering their computational as well as quality metrics at the draft genome level. We divide the quality metrics into three groups: g1 measures the goodness of the assemblies, g2 measures the problems of the assemblies, and g3 measures the conservation elements in the assemblies. Our results demonstrate that the assemblers ABySS and IDBA-UD exhibit a good performance for the studied data from fungal genomes in terms of running time, memory, and quality. The results suggest that whole genome shotgun sequencing projects should make use of different assemblers by considering their merits.

Temporal self-attention for risk prediction from electronic health records using non-stationary kernel approximation.

Effective modeling of patient representation from electronic health records (EHRs) is increasingly becoming a vital research topic. Yet, modeling the non-stationarity in EHR data has received less attention. Most existing studies follow a strong assumption of stationarity in patient representation from EHRs. However, in practice, a patient's visits are irregularly spaced over a relatively long period of time, and disease progression patterns exhibit non-stationarity. Furthermore, the time gaps between patient visits often encapsulate significant domain knowledge, potentially revealing undiscovered patterns that characterize specific medical conditions. To address these challenges, we introduce a new method which combines the self-attention mechanism with non-stationary kernel approximation to capture both contextual information and temporal relationships between patient visits in EHRs. To assess the effectiveness of our proposed approach, we use two real-world EHR datasets, comprising a total of 76,925 patients, for the task of predicting the next diagnosis code for a patient, given their EHR history. The first dataset is a general EHR cohort and consists of 11,451 patients with a total of 3,485 unique diagnosis codes. The second dataset is a disease-specific cohort that includes 65,474 pregnant patients and encompasses a total of 9,782 unique diagnosis codes. Our experimental evaluation involved nine prediction models, categorized into three distinct groups. Group 1 comprises the baselines: original self-attention with positional encoding model, RETAIN model, and LSTM model. Group 2 includes models employing self-attention with stationary kernel approximations, specifically incorporating three variations of Bochner's feature maps. Lastly, Group 3 consists of models utilizing self-attention with non-stationary kernel approximations, including quadratic, cubic, and bi-quadratic polynomials. The experimental results demonstrate that non-stationary kernels significantly outperformed baseline methods for NDCG@10 and Hit@10 metrics in both datasets. The performance boost was more substantial in dataset 1 for the NDCG@10 metric. On the other hand, stationary Kernels showed significant but smaller gains over baselines and were nearly as effective as Non-stationary Kernels for Hit@10 in dataset 2. These findings robustly validate the efficacy of employing non-stationary kernels for temporal modeling of EHR data, and emphasize the importance of modeling non-stationary temporal information in healthcare prediction tasks.

PredictPTB: an interpretable preterm birth prediction model using attention-based recurrent neural networks.

BACKGROUND: Early identification of pregnant women at risk for preterm birth (PTB), a major cause of infant mortality and morbidity, has a significant potential to improve prenatal care. However, we lack effective predictive models which can accurately forecast PTB and complement these predictions with appropriate interpretations for clinicians. In this work, we introduce a clinical prediction model (PredictPTB) which combines variables (medical codes) readily accessible through electronic health record (EHR) to accurately predict the risk of preterm birth at 1, 3, 6, and 9 months prior to delivery. METHODS: The architecture of PredictPTB employs recurrent neural networks (RNNs) to model the longitudinal patient's EHR visits and exploits a single code-level attention mechanism to improve the predictive performance, while providing temporal code-level and visit-level explanations for the prediction results. We compare the performance of different combinations of prediction time-points, data modalities, and data windows. We also present a case-study of our model's interpretability illustrating how clinicians can gain some transparency into the predictions. RESULTS: Leveraging a large cohort of 222,436 deliveries, comprising a total of 27,100 unique clinical concepts, our model was able to predict preterm birth with an ROC-AUC of 0.82, 0.79, 0.78, and PR-AUC of 0.40, 0.31, 0.24, at 1, 3, and 6 months prior to delivery, respectively. Results also confirm that observational data modalities (such as diagnoses) are more predictive for preterm birth than interventional data modalities (e.g., medications and procedures). CONCLUSIONS: Our results demonstrate that PredictPTB can be utilized to achieve accurate and scalable predictions for preterm birth, complemented by explanations that directly highlight evidence in the patient's EHR timeline.

Machine Learning for Healthcare Wearable Devices: The Big Picture.

Using artificial intelligence and machine learning techniques in healthcare applications has been actively researched over the last few years. It holds promising opportunities as it is used to track human activities and vital signs using wearable devices and assist in diseases' diagnosis, and it can play a great role in elderly care and patient's health monitoring and diagnostics. With the great technological advances in medical sensors and miniaturization of electronic chips in the recent five years, more applications are being researched and developed for wearable devices. Despite the remarkable growth of using smart watches and other wearable devices, a few of these massive research efforts for machine learning applications have found their way to market. In this study, a review of the different areas of the recent machine learning research for healthcare wearable devices is presented. Different challenges facing machine learning applications on wearable devices are discussed. Potential solutions from the literature are presented, and areas open for improvement and further research are highlighted.

LocationSpark: In-memory Distributed Spatial Query Processing and Optimization.

Due to the ubiquity of spatial data applications and the large amounts of spatial data that these applications generate and process, there is a pressing need for scalable spatial query processing. In this paper, we present new techniques for spatial query processing and optimization in an in-memory and distributed setup to address scalability. More specifically, we introduce new techniques for handling query skew that commonly happens in practice, and minimizes communication costs accordingly. We propose a distributed query scheduler that uses a new cost model to minimize the cost of spatial query processing. The scheduler generates query execution plans that minimize the effect of query skew. The query scheduler utilizes new spatial indexing techniques based on bitmap filters to forward queries to the appropriate local nodes. Each local computation node is responsible for optimizing and selecting its best local query execution plan based on the indexes and the nature of the spatial queries in that node. All the proposed spatial query processing and optimization techniques are prototyped inside Spark, a distributed memory-based computation system. Our prototype system is termed LocationSpark. The experimental study is based on real datasets and demonstrates that LocationSpark can enhance distributed spatial query processing by up to an order of magnitude over existing in-memory and distributed spatial systems.

FastRNABindR: Fast and Accurate Prediction of Protein-RNA Interface Residues.

A wide range of biological processes, including regulation of gene expression, protein synthesis, and replication and assembly of many viruses are mediated by RNA-protein interactions. However, experimental determination of the structures of protein-RNA complexes is expensive and technically challenging. Hence, a number of computational tools have been developed for predicting protein-RNA interfaces. Some of the state-of-the-art protein-RNA interface predictors rely on position-specific scoring matrix (PSSM)-based encoding of the protein sequences. The computational efforts needed for generating PSSMs severely limits the practical utility of protein-RNA interface prediction servers. In this work, we experiment with two approaches, random sampling and sequence similarity reduction, for extracting a representative reference database of protein sequences from more than 50 million protein sequences in UniRef100. Our results suggest that random sampled databases produce better PSSM profiles (in terms of the number of hits used to generate the profile and the distance of the generated profile to the corresponding profile generated using the entire UniRef100 data as well as the accuracy of the machine learning classifier trained using these profiles). Based on our results, we developed FastRNABindR, an improved version of RNABindR for predicting protein-RNA interface residues using PSSM profiles generated using 1% of the UniRef100 sequences sampled uniformly at random. To the best of our knowledge, FastRNABindR is the only protein-RNA interface residue prediction online server that requires generation of PSSM profiles for query sequences and accepts hundreds of protein sequences per submission. Our approach for determining the optimal BLAST database for a protein-RNA interface residue classification task has the potential of substantially speeding up, and hence increasing the practical utility of, other amino acid sequence based predictors of protein-protein and protein-DNA interfaces.

A Practical and Scalable Tool to Find Overlaps between Sequences.

The evolution of the next generation sequencing technology increases the demand for efficient solutions, in terms of space and time, for several bioinformatics problems. This paper presents a practical and easy-to-implement solution for one of these problems, namely, the all-pairs suffix-prefix problem, using a compact prefix tree. The paper demonstrates an efficient construction of this time-efficient and space-economical tree data structure. The paper presents techniques for parallel implementations of the proposed solution. Experimental evaluation indicates superior results in terms of space and time over existing solutions. Results also show that the proposed technique is highly scalable in a parallel execution environment.

Using the Sadakane compressed suffix tree to solve the all-pairs suffix-prefix problem.

The all-pairs suffix-prefix matching problem is a basic problem in string processing. It has an application in the de novo genome assembly task, which is one of the major bioinformatics problems. Due to the large size of the input data, it is crucial to use fast and space efficient solutions. In this paper, we present a space-economical solution to this problem using the generalized Sadakane compressed suffix tree. Furthermore, we present a parallel algorithm to provide more speed for shared memory computers. Our sequential and parallel algorithms are optimized by exploiting features of the Sadakane compressed index data structure. Experimental results show that our solution based on the Sadakane's compressed index consumes significantly less space than the ones based on noncompressed data structures like the suffix tree and the enhanced suffix array. Our experimental results show that our parallel algorithm is efficient and scales well with increasing number of processors.