Multi-Task Learning (MTL) is a growing subject of interest in deep learning, due to its ability to train models more efficiently on multiple tasks compared to using a group of conventional single-task models. However, MTL can be impractical as certain tasks can dominate training and hurt performance in others, thus making some tasks perform better in a single-task model compared to a multi-task one. Such problems are broadly classified as negative transfer, and many prior approaches in the literature have been made to mitigate these issues. One such current approach to alleviate negative transfer is to weight each of the losses so that they are on the same scale. Whereas current loss balancing approaches rely on either optimization or complex numerical analysis, none directly scale the losses based on their observed magnitudes. We propose multiple techniques for loss balancing based on scaling by the exponential moving average and benchmark them against current best-performing methods on three established datasets. On these datasets, they achieve comparable, if not higher, performance compared to current best-performing methods.

BACKGROUND: Sequencing partial 16S rRNA genes is a cost effective method for quantifying the microbial composition of an environment, such as the human gut. However, downstream analysis relies on binning reads into microbial groups by either considering each unique sequence as a different microbe, querying a database to get taxonomic labels from sequences, or clustering similar sequences together. However, these approaches do not fully capture evolutionary relationships between microbes, limiting the ability to identify differentially abundant groups of microbes between a diseased and control cohort. We present sequence-based biomarkers (SBBs), an aggregation method that groups and aggregates microbes using single variants and combinations of variants within their 16S sequences. We compare SBBs against other existing aggregation methods (OTU clustering and Microphenoor DiTaxa features) in several benchmarking tasks: biomarker discovery via permutation test, biomarker discovery via linear discriminant analysis, and phenotype prediction power. We demonstrate the SBBs perform on-par or better than the state-of-the-art methods in biomarker discovery and phenotype prediction. RESULTS: On two independent datasets, SBBs identify differentially abundant groups of microbes with similar or higher statistical significance than existing methods in both a permutation-test-based analysis and using linear discriminant analysis effect size. . By grouping microbes by SBB, we can identify several differentially abundant microbial groups (FDR <.1) between children with autism and neurotypical controls in a set of 115 discordant siblings. Porphyromonadaceae, Ruminococcaceae, and an unnamed species of Blastocystis were significantly enriched in autism, while Veillonellaceae was significantly depleted. Likewise, aggregating microbes by SBB on a dataset of obese and lean twins, we find several significantly differentially abundant microbial groups (FDR<.1). We observed Megasphaera andSutterellaceae highly enriched in obesity, and Phocaeicola significantly depleted. SBBs also perform on bar with or better than existing aggregation methods as features in a phenotype prediction model, predicting the autism phenotype with an ROC-AUC score of .64 and the obesity phenotype with an ROC-AUC score of .84. CONCLUSIONS: SBBs provide a powerful method for aggregating microbes to perform differential abundance analysis as well as phenotype prediction. Our source code can be freely downloaded from http://github.com/briannachrisman/16s_biomarkers .

BACKGROUND: Complex human health conditions with etiological heterogeneity like Autism Spectrum Disorder (ASD) often pose a challenge for traditional genome-wide association study approaches in defining a clear genotype to phenotype model. Coalitional game theory (CGT) is an exciting method that can consider the combinatorial effect of groups of variants working in concert to produce a phenotype. CGT has been applied to associate likely-gene-disrupting variants encoded from whole genome sequence data to ASD; however, this previous approach cannot take into account for prior biological knowledge. Here we extend CGT to incorporate a priori knowledge from biological networks through a game theoretic centrality measure based on Shapley value to rank genes by their relevance-the individual gene's synergistic influence in a gene-to-gene interaction network. Game theoretic centrality extends the notion of Shapley value to the evaluation of a gene's contribution to the overall connectivity of its corresponding node in a biological network. RESULTS: We implemented and applied game theoretic centrality to rank genes on whole genomes from 756 multiplex autism families. Top ranking genes with the highest game theoretic centrality in both the weighted and unweighted approaches were enriched for pathways previously associated with autism, including pathways of the immune system. Four of the selected genes HLA-A, HLA-B, HLA-G, and HLA-DRB1-have also been implicated in ASD and further support the link between ASD and the human leukocyte antigen complex. CONCLUSIONS: Game theoretic centrality can prioritize influential, disease-associated genes within biological networks, and assist in the decoding of polygenic associations to complex disorders like autism.

In this paper, we describe challenges in the development of a mobile charades-style game for delivery of social training to children with Autism Spectrum Disorder (ASD). Providing real-time feedback and adapting game difficulty in response to the child's performance necessitates the integration of emotion classifiers into the system. Due to the limited performance of existing emotion recognition platforms for children with ASD, we propose a novel technique to automatically extract emotion-labeled frames from video acquired from game sessions, which we hypothesize can be used to train new emotion classifiers to overcome these limitations. Our technique, which uses probability scores from three different classifiers and meta information from game sessions, correctly identified 83% of frames compared to a baseline of 51.6% from the best emotion classification API evaluated in our work.

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by repetitive behaviors, narrow interests, and deficits in social interaction and communication ability. An increasing emphasis is being placed on the development of innovative digital and mobile systems for their potential in therapeutic applications outside of clinical environments. Due to recent advances in the field of computer vision, various emotion classifiers have been developed, which have potential to play a significant role in mobile screening and therapy for developmental delays that impair emotion recognition and expression. However, these classifiers are trained on datasets of predominantly neurotypical adults and can sometimes fail to generalize to children with autism. The need to improve existing classifiers and develop new systems that overcome these limitations necessitates novel methods to crowdsource labeled emotion data from children. In this paper, we present a mobile charades-style game, Guess What?, from which we derive egocentric video with a high density of varied emotion from a 90-second game session. We then present a framework for semi-automatic labeled frame extraction from these videos using meta information from the game session coupled with classification confidence scores. Results show that 94%, 81%, 92%, and 56% of frames were automatically labeled correctly for categories disgust, neutral, surprise, and scared respectively, though performance for angry and happy did not improve significantly from the baseline.

Background: Existing screening tools for early detection of autism are expensive, cumbersome, time- intensive, and sometimes fall short in predictive value. In this work, we sought to apply Machine Learning (ML) to gold standard clinical data obtained across thousands of children at-risk for autism spectrum disorder to create a low-cost, quick, and easy to apply autism screening tool. Methods: Two algorithms are trained to identify autism, one based on short, structured parent-reported questionnaires and the other on tagging key behaviors from short, semi-structured home videos of children. A combination algorithm is then used to combine the results into a single assessment of higher accuracy. To overcome the scarcity, sparsity, and imbalance of training data, we apply novel feature selection, feature engineering, and feature encoding techniques. We allow for inconclusive determination where appropriate in order to boost screening accuracy when conclusive. The performance is then validated in a controlled clinical study. Results: A multi-center clinical study of n = 162 children is performed to ascertain the performance of these algorithms and their combination. We demonstrate a significant accuracy improvement over standard screening tools in measurements of AUC, sensitivity, and specificity. Conclusion: These findings suggest that a mobile, machine learning process is a reliable method for detection of autism outside of clinical settings. A variety of confounding factors in the clinical analysis are discussed along with the solutions engineered into the algorithms. Final results are statistically limited and will benefit from future clinical studies to extend the sample size.

Existing screening tools for early detection of autism are expensive, cumbersome, time-intensive, and sometimes fall short in predictive value. In this work, we apply Machine Learning to gold standard clinical data obtained across thousands of children at risk for autism spectrum disorders to create a low-cost, quick, and easy to apply autism screening tool that performs better than most widely used standardized instruments. This new tool combines two screening methods into a single assessment, one based on short, structured parent-reported questionnaires and the other based on tagging key behaviors from short, semi-structured home videos of children. We demonstrate a significant accuracy improvement over standard screening tools in a clinical study sample of 162 children. We further discuss the challenge of extending machine learning algorithms to conditions beyond autism, and we propose a generalized framework for using machine learning algorithms to simultaneously search for the presence of many different conditions.

BACKGROUND: Next Generation Genome sequencing techniques became affordable for massive sequencing efforts devoted to clinical characterization of human diseases. However, the cost of providing cloud-based data analysis of the mounting datasets remains a concerning bottleneck for providing cost-effective clinical services. To address this computational problem, it is important to optimize the variant analysis workflow and the used analysis tools to reduce the overall computational processing time, and concomitantly reduce the processing cost. Furthermore, it is important to capitalize on the use of the recent development in the cloud computing market, which have witnessed more providers competing in terms of products and prices. RESULTS: In this paper, we present a new package called MC-GenomeKey (Multi-Cloud GenomeKey) that efficiently executes the variant analysis workflow for detecting and annotating mutations using cloud resources from different commercial cloud providers. Our package supports Amazon, Google, and Azure clouds, as well as, any other cloud platform based on OpenStack. Our package allows different scenarios of execution with different levels of sophistication, up to the one where a workflow can be executed using a cluster whose nodes come from different clouds. MC-GenomeKey also supports scenarios to exploit the spot instance model of Amazon in combination with the use of other cloud platforms to provide significant cost reduction. To the best of our knowledge, this is the first solution that optimizes the execution of the workflow using computational resources from different cloud providers. CONCLUSIONS: MC-GenomeKey provides an efficient multicloud based solution to detect and annotate mutations. The package can run in different commercial cloud platforms, which enables the user to seize the best offers. The package also provides a reliable means to make use of the low-cost spot instance model of Amazon, as it provides an efficient solution to the sudden termination of spot machines as a result of a sudden price increase. The package has a web-interface and it is available for free for academic use.

Methods for automated facial expression recognition - identifying faces as happy, sad, angry, etc. - typically rely on the classification of features extracted from images. These features, designed to encode shape and texture information, depend on both (1) the expression an individual is making, and (2) the individual's physical characteristics and lighting conditions of the image. To reduce the effect of (2), a common strategy is to establish a "baseline" for an individual and subtract out this individual's baseline neutral feature. This extra neutral feature information often is not available - in particular for in-the-wild, real-time classification of a previously unseen subject. Thus, in order to implement "neutral subtraction," one must estimate the individual's neutral feature. Existing methods to do this are susceptible to class imbalance at test time (e.g., averaging over all facial features), require a more complex model specific to the individual to be trained, or are restricted to features computed entirely from tracked landmark points (taking advantage of a subset of "stable points" which move little as an individual emotes). We extend neutral subtraction to different computer vision feature spaces as a method to correct for inter-face and lighting variance. We further propose a simple, real-time method which is robust to class imbalance and in principal works over a wide class of feature choices. We test this method on feature extraction techniques that lead to high baseline accuracy without neutral subtraction (97% on the Extended Cohn-Kanade Dataset). We find that on difficult classification tasks our method recovers almost 2/3 of the ~ 8% gain shown by a "cheating" neutral-subtracted feature classifier, which uses examples that have been labeled as neutral, validating with both HOG and SIFT features.

The diabetes healthcare provider plays a key role in interpreting blood glucose trends, but few institutions have successfully integrated patient home glucose data in the electronic health record (EHR). Published implementations to date have required custom interfaces, which limit wide-scale replication. We piloted automated integration of continuous glucose monitor data in the EHR using widely available consumer technology for 10 pediatric patients with insulin-dependent diabetes. Establishment of a passive data communication bridge via a patient's/parent's smartphone enabled automated integration and analytics of patient device data within the EHR between scheduled clinic visits. It is feasible to utilize available consumer technology to assess and triage home diabetes device data within the EHR, and to engage patients/parents and improve healthcare provider workflow.

The dramatic fall of next generation sequencing (NGS) cost in recent years positions the price in range of typical medical testing, and thus whole genome analysis (WGA) may be a viable clinical diagnostic tool. Modern sequencing platforms routinely generate petabyte data. The current challenge lies in calling and analyzing this large-scale data, which has become the new time and cost rate-limiting step. To address the computational limitations and optimize the cost, we have developed COSMOS ( http://cosmos.hms.harvard.edu ) , a scalable, parallelizable workflow management system running on clouds (e.g., Amazon Web Services or Google Clouds). Using COSMOS [ 1 ], we have constructed a NGS analysis pipeline implementing the Genome Analysis Toolkit - GATK v3.1 - best practice protocol [ 2 , 3 ], a widely accepted industry standard developed by the Broad Institute. COSMOS performs a thorough sequence analysis, including quality control, alignment, variant calling and an unprecedented level of annotation using a custom extension of ANNOVAR. COSMOS takes advantage of parallelization and the resources of a high-performance compute cluster, either local or in the cloud, to process datasets of up to the petabyte scale, which is becoming standard in NGS. This approach enables the timely and cost-effective implementation of NGS analysis, allowing for it to be used in a clinical setting and translational medicine. With COSMOS we reduced the whole genome data analysis cost under the $100 barrier, placing it within a reimbursable cost point and in clinical time , providing a significant change to the landscape of genomic analysis and cement the utility of cloud environment as a resource for Petabyte-scale genomic research.

SUMMARY: Efficient workflows to shepherd clinically generated genomic data through the multiple stages of a next-generation sequencing pipeline are of critical importance in translational biomedical science. Here we present COSMOS, a Python library for workflow management that allows formal description of pipelines and partitioning of jobs. In addition, it includes a user interface for tracking the progress of jobs, abstraction of the queuing system and fine-grained control over the workflow. Workflows can be created on traditional computing clusters as well as cloud-based services. AVAILABILITY AND IMPLEMENTATION: Source code is available for academic non-commercial research purposes. Links to code and documentation are provided at http://lpm.hms.harvard.edu and http://wall-lab.stanford.edu. CONTACT: [email protected] or [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

OBJECTIVE: To extract disorder-associated genes from the scientific literature in PubMed with greater sensitivity for literature-based support than existing methods. METHODS: We developed a PubMed query to retrieve disorder-related, original research articles. Then we applied a rule-based text-mining algorithm with keyword matching to extract target disorders, genes with significant results, and the type of study described by the article. RESULTS: We compared our resulting candidate disorder genes and supporting references with existing databases. We demonstrated that our candidate gene set covers nearly all genes in manually curated databases, and that the references supporting the disorder-gene link are more extensive and accurate than other general purpose gene-to-disorder association databases. CONCLUSIONS: We implemented a novel publication search tool to find target articles, specifically focused on links between disorders and genotypes. Through comparison against gold-standard manually updated gene-disorder databases and comparison with automated databases of similar functionality we show that our tool can search through the entirety of PubMed to extract the main gene findings for human diseases rapidly and accurately.

UNLABELLED: Roundup is an online database of gene orthologs for over 1800 genomes, including 226 Eukaryota, 1447 Bacteria, 113 Archaea and 21 Viruses. Orthologs are inferred using the Reciprocal Smallest Distance algorithm. Users may query Roundup for single-linkage clusters of orthologous genes based on any group of genomes. Annotated query results may be viewed in a variety of ways including as clusters of orthologs and as phylogenetic profiles. Genomic results may be downloaded in formats suitable for functional as well as phylogenetic analysis, including the recent OrthoXML standard. In addition, gene IDs can be retrieved using FASTA sequence search. All source code and orthologs are freely available. AVAILABILITY: http://roundup.hms.harvard.edu.

BACKGROUND: Bioinformatics services have been traditionally provided in the form of a web-server that is hosted at institutional infrastructure and serves multiple users. This model, however, is not flexible enough to cope with the increasing number of users, increasing data size, and new requirements in terms of speed and availability of service. The advent of cloud computing suggests a new service model that provides an efficient solution to these problems, based on the concepts of "resources-on-demand" and "pay-as-you-go". However, cloud computing has not yet been introduced within bioinformatics servers due to the lack of usage scenarios and software layers that address the requirements of the bioinformatics domain. RESULTS: In this paper, we provide different use case scenarios for providing cloud computing based services, considering both the technical and financial aspects of the cloud computing service model. These scenarios are for individual users seeking computational power as well as bioinformatics service providers aiming at provision of personalized bioinformatics services to their users. We also present elasticHPC, a software package and a library that facilitates the use of high performance cloud computing resources in general and the implementation of the suggested bioinformatics scenarios in particular. Concrete examples that demonstrate the suggested use case scenarios with whole bioinformatics servers and major sequence analysis tools like BLAST are presented. Experimental results with large datasets are also included to show the advantages of the cloud model. CONCLUSIONS: Our use case scenarios and the elasticHPC package are steps towards the provision of cloud based bioinformatics services, which would help in overcoming the data challenge of recent biological research. All resources related to elasticHPC and its web-interface are available at http://www.elasticHPC.org.

SUMMARY: We developed a package TripletSearch to compute relationships within triplets of genes based on Roundup, an orthologous gene database containing >1500 genomes. These relationships, derived from the coevolution of genes, provide valuable information in the detection of biological network organization from the local to the system level, in the inference of protein functions and in the identification of functional orthologs. To run the computation, users need to provide the GI IDs of the genes of interest. AVAILABILITY: http://wall.hms.harvard.edu/sites/default/files/tripletSearch.tar.gz CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: A "phylogenetic profile" refers to the presence or absence of a gene across a set of organisms, and it has been proven valuable for understanding gene functional relationships and network organization. Despite this success, few studies have attempted to search beyond just pairwise relationships among genes. Here we search for logic relationships involving three genes, and explore its potential application in gene network analyses. RESULTS: Taking advantage of a phylogenetic matrix constructed from the large orthologs database Roundup, we invented a method to create balanced profiles for individual triplets of genes that guarantee equal weight on the different phylogenetic scenarios of coevolution between genes. When we applied this idea to LAPP, the method to search for logic triplets of genes, the balanced profiles resulted in significant performance improvement and the discovery of hundreds of thousands more putative triplets than unadjusted profiles. We found that logic triplets detected biological network organization and identified key proteins and their functions, ranging from neighbouring proteins in local pathways, to well separated proteins in the whole pathway, and to the interactions among different pathways at the system level. Finally, our case study suggested that the directionality in a logic relationship and the profile of a triplet could disclose the connectivity between the triplet and surrounding networks. CONCLUSION: Balanced profiles are superior to the raw profiles employed by traditional methods of phylogenetic profiling in searching for high order gene sets. Gene triplets can provide valuable information in detection of biological network organization and identification of key genes at different levels of cellular interaction.

In this overview to biomedical computing in the cloud, we discussed two primary ways to use the cloud (a single instance or cluster), provided a detailed example using NGS mapping, and highlighted the associated costs. While many users new to the cloud may assume that entry is as straightforward as uploading an application and selecting an instance type and storage options, we illustrated that there is substantial up-front effort required before an application can make full use of the cloud's vast resources. Our intention was to provide a set of best practices and to illustrate how those apply to a typical application pipeline for biomedical informatics, but also general enough for extrapolation to other types of computational problems. Our mapping example was intended to illustrate how to develop a scalable project and not to compare and contrast alignment algorithms for read mapping and genome assembly. Indeed, with a newer aligner such as Bowtie, it is possible to map the entire African genome using one m2.2xlarge instance in 48 hours for a total cost of approximately $48 in computation time. In our example, we were not concerned with data transfer rates, which are heavily influenced by the amount of available bandwidth, connection latency, and network availability. When transferring large amounts of data to the cloud, bandwidth limitations can be a major bottleneck, and in some cases it is more efficient to simply mail a storage device containing the data to AWS (http://aws.amazon.com/importexport/). More information about cloud computing, detailed cost analysis, and security can be found in references.

BACKGROUND: Large comparative genomics studies and tools are becoming increasingly more compute-expensive as the number of available genome sequences continues to rise. The capacity and cost of local computing infrastructures are likely to become prohibitive with the increase, especially as the breadth of questions continues to rise. Alternative computing architectures, in particular cloud computing environments, may help alleviate this increasing pressure and enable fast, large-scale, and cost-effective comparative genomics strategies going forward. To test this, we redesigned a typical comparative genomics algorithm, the reciprocal smallest distance algorithm (RSD), to run within Amazon's Elastic Computing Cloud (EC2). We then employed the RSD-cloud for ortholog calculations across a wide selection of fully sequenced genomes. RESULTS: We ran more than 300,000 RSD-cloud processes within the EC2. These jobs were farmed simultaneously to 100 high capacity compute nodes using the Amazon Web Service Elastic Map Reduce and included a wide mix of large and small genomes. The total computation time took just under 70 hours and cost a total of $6,302 USD. CONCLUSIONS: The effort to transform existing comparative genomics algorithms from local compute infrastructures is not trivial. However, the speed and flexibility of cloud computing environments provides a substantial boost with manageable cost. The procedure designed to transform the RSD algorithm into a cloud-ready application is readily adaptable to similar comparative genomics problems.

Agglomerating results from studies of individual biological components has shown the potential to produce biomedical discovery and the promise of therapeutic development. Such knowledge integration could be tremendously facilitated by automated text mining for relation extraction in the biomedical literature. Relation extraction systems cannot be developed without substantial datasets annotated with ground truth for benchmarking and training. The creation of such datasets is hampered by the absence of a resource for launching a distributed annotation effort, as well as by the lack of a standardized annotation schema. We have developed an annotation schema and an annotation tool which can be widely adopted so that the resulting annotated corpora from a multitude of disease studies could be assembled into a unified benchmark dataset. The contribution of this paper is threefold. First, we provide an overview of available benchmark corpora and derive a simple annotation schema for specific binary relation extraction problems such as protein-protein and gene-disease relation extraction. Second, we present BioNotate: an open source annotation resource for the distributed creation of a large corpus. Third, we present and make available the results of a pilot annotation effort of the autism disease network.

SUMMARY: We have created a tool for ortholog and phylogenetic profile retrieval called Roundup. Roundup is backed by a massive repository of orthologs and associated evolutionary distances that was built using the reciprocal smallest distance algorithm, an approach that has been shown to improve upon alternative approaches of ortholog detection, such as reciprocal blast. Presently, the Roundup repository contains all possible pair-wise comparisons for over 250 genomes, including 32 Eukaryotes, more than doubling the coverage of any similar resource. The orthologs are accessible through an intuitive web interface that allows searches by genome or gene identifier, presenting results as phylogenetic profiles together with gene and molecular function annotations. Results may be downloaded as phylogenetic matrices for subsequent analysis, including the construction of whole-genome phylogenies based on gene-content data. AVAILABILITY: http://rodeo.med.harvard.edu/tools/roundup.

Abstract Summary: We developed an algorithm that improves upon the common procedure of taking reciprocal best blast hits (rbh) in the identification of orthologs. The method—reciprocal smallest distance algorithm (rsd)—relies on global sequence alignment and maximum likelihood estimation of evolutionary distances to detect orthologs between two genomes. rsd finds many putative orthologs missed by rbh because it is less likely than rbh to be misled by the presence of a close paralog. Availability: A Python program and ReadMe file are freely available from: http://charles.stanford.edu/~dennis/research.html