A major challenge in working with single-cell RNA sequencing data is the prevalence of "dropout," when some transcripts' expression values are erroneously not captured. Addressing this issue, which produces zero-inflated count data, is crucial for many downstream data analyses including the inference of gene regulatory networks (GRNs). In this paper, we introduce two novel contributions. First, we propose Dropout Augmentation (DA), a simple but effective model regularization method to improve resilience to zero inflation in single-cell data by augmenting the data with synthetic dropout events. DA offers a new perspective to solve the "dropout" problem beyond imputation. Second, we present DAZZLE, a stabilized and robust version of the autoencoder-based structure equation model for GRN inference using the DA concept. Benchmark experiments illustrate the improved performance and increased stability of the proposed DAZZLE model over existing approaches. The practical application of the DAZZLE model on a longitudinal mouse microglia dataset containing over 15,000 genes illustrates its ability to handle real-world single cell data with minimal gene filtration. The improved robustness and stability of DAZZLE make it a practical and valuable addition to the toolkit for GRN inference from single-cell data. Finally, we propose that Dropout Augmentation may have wider applications beyond the GRN-inference problem. Project website: https://bcb.cs.tufts.edu/DAZZLE.

This has been an unusual year, even for ISMB. Due to the worldwide COVID-19 pandemic, the annual ISMB meeting (the 28th Annual Conference on Intelligent Systems for Molecular Biology), initially intended to be held in Montreal, Canada, was instead run as a fully virtual conference on July 13–16, 2020. A huge debt of gratitude is due to International Society for Computational Biology (ISCB) Executive Director Diane Kovats, Director of Conferences Steven Leard and the rest of the conference team for arranging to move the entire meeting online on extremely short notice! ISMB is the flagship conference of the ISCB and may be considered the world’s premier forum for dissemination of scientific research in computational biology. The Proceedings Track at ISMB provides an opportunity for members of the computational biology research community to submit full length papers featuring original research for thorough review. Both theoretical and applied submissions are welcomed in any field of computational biology; accepted papers appear in this special issue of Bioinformatics and are presented at the conference. The review process this year was managed by dividing the submitted manuscripts among nine scientific review areas. Program committee recruitment and reviewing were overseen by the Senior Program Committee (SPC), which includes the Proceedings Chairs and the nine sets of Area Chairs (AC), several of whom were nominated by the Communities of Special Interests (COSIs). Reviewers looked for novel, important biological insights; correctness; clarity and potential impact. Once reviews were submitted, papers with discrepant scores were discussed by the reviewers and the relevant ACs, often resulting in refined reviews and scores. Final acceptance decisions were made by the entire SPC, including the Proceedings Chairs listed as authors of this piece, plus the 18 ACs listed in Table 1. Thematic areas of ISMB2020 Notes: The table lists the ACs for each theme, the number of reviewed papers, the number of accepted papers and the acceptance rate for each area. The special area of General Computational Biology was designated to handle papers on emerging topics or those not fitting in any of the other areas. Thematic areas of ISMB2020 Notes: The table lists the ACs for each theme, the number of reviewed papers, the number of accepted papers and the acceptance rate for each area. The special area of General Computational Biology was designated to handle papers on emerging topics or those not fitting in any of the other areas. Eight of the reviewing areas in Table 1 focused on particular classes of biological problems, some of which are quite broad. The ninth area, General Computational Biology, is intended for submissions on emerging topics or those that do not fit well in other reviewing areas. Some more common topics of manuscripts handled by this area in 2020 included ontologies and bioNLP, single-cell methodologies, CRISPR and imaging applications. Some papers were moved between reviewing areas to avoid conflicts of interest. Abstracts, previously published papers, position papers, perspectives and reviews were not eligible for submission to the ISMB Proceedings track. In total, 329 manuscripts were submitted and sent out for review. We wish to especially acknowledge all 226 program committee members and 264 sub-reviewers for doing a tough job under adverse circumstances. Many of the world’s initial COVID-19 shutdown periods overlapped with the final weeks of the review period, yet the deadlines for reviewing and decision-making were met. Overall, we received a total of 1078 reviews, with an average of 3.3 reviews per submission. The distribution of the number of reviews per paper is shown in Figure 1. Nearly all manuscripts received at least three reviews; the average was 3.3 Conditional acceptance information was provided to the authors a month after the submission deadline and final acceptance of revised manuscripts was determined a month later. Overall, 65 manuscripts were accepted, for a final acceptance rate of 20%. The distribution of papers reviewed under the different thematic areas is shown in Table 1. The authors had the opportunity to designate the particular COSI session for which they believed their paper was most suitable. However, many of the submitted manuscripts made no specific COSI session request. Accordingly, accepted papers were first offered to the author-requested COSI if there was one, but the final assignments of accepted papers to COSI sessions were made by the COSI and Proceedings Chairs on the basis of both author and COSI requests. Three papers were assigned to the General Computational Biology session for presentation. The ultimate distribution of accepted papers to COSIs is shown in Table 2. COSI distribution of accepted ISMB 2020 Proceedings papers COSI distribution of accepted ISMB 2020 Proceedings papers We are grateful to the ISMB 2020 Steering Committee, Diane Kovats, Steven Leard, Pat Rodenburg and Seth Mulholland for their support, guidance and cheerful handling of both minor logistical questions and big existential ones at this difficult time. Special thanks go also to the SPC, the reviewers and their sub-reviewers, for going above and beyond in maintaining the high standards of ISMB on a time scale that would have been tight even under typical conditions. Thanks to the COSI chairs for great AC nominations, the ACs for being exceptionally responsive and professional and the COSI contacts for additional help during the process of identifying reviewers and in incorporating accepted papers into their programs. One of the bright spots in recent months has been the way the entire international research community has come together to focus on a common challenge. Data and code sharing have increased, journals have become more willing to consider manuscripts released as preprints, paywalls have been loosened and new collaborations have arisen. We see the ISMB community reflecting these trends, and we are heartened by the hope that some of these moves toward more open science will continue. Finally, we want to thank everyone who has participated in this conference–from keynote speakers to students attending for the first time. You make this meeting, and the whole ISMB community, what it is. We hereby invite you to read the papers in this Proceedings volume. Have a safe, healthy year and we hope to be able to welcome you in person to ISMB 2021 in Lyon, France.

Research in computational biology has given rise to a vast number of methods developed to solve scientific problems.For areas in which many approaches exist, researchers have a hard time deciding which tool to select to address a scientific challenge, as essentially all publications introducing a new method will claim better performance than all others.Not all of these claims can be correct.Equally, for this same reason, developers struggle to demonstrate convincingly that they created a new and superior algorithm or implementation.Moreover, the developer community often has difficulty discerning which new approaches constitute true scientific advances for the field.The obvious answer to this conundrum is to develop benchmarks-meaning standard points of reference that facilitate evaluating the performance of different tools-allowing both users and developers to compare multiple tools in an unbiased fashion.Broadly speaking, benchmarks consist of input data that methods are meant to operate upon, expected output data against which tool output can be compared, a specification of metrics used to assess performance, and performance values of sets of tools that have been run through the benchmark.Developing good and comprehensive benchmarks, in which the performance metrics of each tool reflect its real-world utility, requires a significant effort.For highly competitive and established fields, such as protein structure predictions, community experiments evaluating the methods have been held periodically to provide blinded assessments of prediction performance.These blinded assessments are perhaps the gold standard on how benchmarks should be run.However, in most areas of computational biology, no such regular blinded contests are available.Instead, many tool developers end up generating their own benchmarks, which they publish alongside a newly developed tool to show its improved performance.The downside of this approach is that, if a new approach is developed in parallel to assembly of the benchmark on which it is evaluated, there is a strong selection bias encouraging the authors to report tool development approaches performing well against the benchmark compared to previous tools.This reporting bias makes most benchmarks that accompany newly developed tools questionable.Even if the authors are aware of this problem and take conscious steps to separate

The articles in this special section were presented at the fourth ACM Conference on Bioinformatics, Computational Biology, and BiomedicalInformatics, held in Washington D.C. in September 2013.

BACKGROUND: Recent increases in genomic studies of the developing human fetus and neonate have led to a need for widespread characterization of the functional roles of genes at different developmental stages. The Gene Ontology (GO), a valuable and widely-used resource for characterizing gene function, offers perhaps the most suitable functional annotation system for this purpose. However, due in part to the difficulty of studying molecular genetic effects in humans, even the current collection of comprehensive GO annotations for human genes and gene products often lacks adequate developmental context for scientists wishing to study gene function in the human fetus. DESCRIPTION: The Developmental FunctionaL Annotation at Tufts (DFLAT) project aims to improve the quality of analyses of fetal gene expression and regulation by curating human fetal gene functions using both manual and semi-automated GO procedures. Eligible annotations are then contributed to the GO database and included in GO releases of human data. DFLAT has produced a considerable body of functional annotation that we demonstrate provides valuable information about developmental genomics. A collection of gene sets (genes implicated in the same function or biological process), made by combining existing GO annotations with the 13,344 new DFLAT annotations, is available for use in novel analyses. Gene set analyses of expression in several data sets, including amniotic fluid RNA from fetuses with trisomies 21 and 18, umbilical cord blood, and blood from newborns with bronchopulmonary dysplasia, were conducted both with and without the DFLAT annotation. CONCLUSIONS: Functional analysis of expression data using the DFLAT annotation increases the number of implicated gene sets, reflecting the DFLAT's improved representation of current knowledge. Blinded literature review supports the validity of newly significant findings obtained with the DFLAT annotations. Newly implicated significant gene sets also suggest specific hypotheses for future research. Overall, the DFLAT project contributes new functional annotation and gene sets likely to enhance our ability to interpret genomic studies of human fetal and neonatal development.

Identifying molecular connections between developmental processes and disease can lead to new hypotheses about health risks at all stages of life. Here we introduce a new approach to identifying significant connections between gene sets and disease genes, and apply it to several gene sets related to human development. To overcome the limits of incomplete and imperfect information linking genes to disease, we pool genes within disease subtrees in the MeSH taxonomy, and we demonstrate that such pooling improves the power and accuracy of our approach. Significance is assessed through permutation. We created a web-based visualization tool to facilitate multi-scale exploration of this large collection of significant connections (http://gda.cs.tufts.edu/development). High-level analysis of the results reveals expected connections between tissue-specific developmental processes and diseases linked to those tissues, and widespread connections to developmental disorders and cancers. Yet interesting new hypotheses may be derived from examining the unexpected connections. We highlight and discuss the implications of three such connections, linking dementia with bone development, polycystic ovary syndrome with cardiovascular development, and retinopathy of prematurity with lung development. Our results provide additional evidence that TGFB lays a key role in the early pathogenesis of polycystic ovary syndrome. Our evidence also suggests that the VEGF pathway and downstream NFKB signaling may explain the complex relationship between bronchopulmonary dysplasia and retinopathy of prematurity, and may form a bridge between two currently-competing hypotheses about the molecular origins of bronchopulmonary dysplasia. Further data exploration and similar queries about other gene sets may generate a variety of new information about the molecular relationships between additional diseases.

MOTIVATION: With the growing availability of high-throughput protein-protein interaction (PPI) data, it has become possible to consider how a protein's local or global network characteristics predict its function. RESULTS: We introduce a graph-theoretic approach that identifies key regulatory proteins in an organism by analyzing proteins' local PPI network structure. We apply the method to the yeast genome and describe several properties of the resulting set of regulatory hubs. Finally, we demonstrate how the identified hubs and putative target gene sets can be used to identify causative, functional regulators of differential gene expression linked to human disease. AVAILABILITY: Code is available at http://bcb.cs.tufts.edu/hubcomps. CONTACT: [email protected]; [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

We present a new approach to semi-supervised anomaly detection. Given a set of training examples believed to come from the same distribution or class, the task is to learn a model that will be able to distinguish examples in the future that do not belong to the same class. Traditional approaches typically compare the position of a new data point to the set of "normal" training data points in a chosen representation of the feature space. For some data sets, the normal data may not have discernible positions in feature space, but do have consistent relationships among some features that fail to appear in the anomalous examples. Our approach learns to predict the values of training set features from the values of other features. After we have formed an ensemble of predictors, we apply this ensemble to new data points. To combine the contribution of each predictor in our ensemble, we have developed a novel, information-theoretic anomaly measure that our experimental results show selects against noisy and irrelevant features. Our results on 47 data sets show that for most data sets, this approach significantly improves performance over current state-of-the-art feature space distance and density-based approaches.

Dynamic molecular interactions play a central role in regulating the functioning of cells and organisms. The availability of experimentally determined large-scale cellular networks, along with other high-throughput experimental data sets that provide snapshots of biological systems at different times and conditions, is increasingly helpful in elucidating interaction dynamics. Here we review the beginnings of a new subfield within computational biology, one focused on the global inference and analysis of the dynamic interactome. This burgeoning research area, which entails a shift from static to dynamic network analysis, promises to be a major step forward in our ability to model and reason about cellular function and behavior.

Gene expression microarrays provide a snapshot of all the transcriptional activity in a biological sample. Unlike most traditional molecular biology tools, which generally allow the study of a single gene or a small set of genes, microarrays facilitate the discovery of totally novel and unexpected functional roles of genes. The power of these tools has been applied to a range of applications, including discovering novel disease subtypes, developing new diagnostic tools, and identifying underlying mechanisms of disease or drug response. However, this technology necessarily produces a large amount of data, challenging us to interpret it by exploiting modern computational and statistical tools. In this brief review, we aim to indicate the major issues involved in microarray analysis and provide a useful starting point for new microarray users. Figure 1 outlines the steps in a typical expression microarray experiment and maps them to the different sections of this review. Figure 1 Overview of steps in a typical gene expression microarray experiment.

We propose a new method for collecting information on regulatory elements found by any motif discovery program. We suggest that combining the results of n leave-one-out motif discovery runs provides additional information. By examining motifs found in n - 1 of the sequences and scoring them on the remaining sequence, we overcome some of the issues arising from noisy data to identify more high-quality motifs. We describe preliminary investigations of this approach, using MEME for motif discovery. We show that the leave-one-out method highlights different motifs than a single MEME run would. We demonstrate that our method increases the power of small datasets. We also explore how the information gain of the method changes as the number of sequences increases. Our approach may be generalized to any number of sequences, and may be applied with any motif-inference package that generates a final population of solutions and scores

Classification of patient samples is a crucial aspect of cancer diagnosis and treatment. We present a method for classifying samples by computational analysis of gene expression data. We consider the classification problem in two parts: class discovery and class prediction. Class discovery refers to the process of dividing samples into reproducible classes that have similar behavior or properties, while class prediction places new samples into already known classes. We describe a method for performing class prediction and illustrate its strength by correctly classifying bone marrow and blood samples from acute leukemia patients. We also describe how to use our predictor to validate newly discovered classes, and we demonstrate how this technique could have discovered the key distinctions among leukemias if they were not already known. This proof-of-concept experiment paves the way for a wealth of future work on the molecular classification and understanding of disease.

Genome maps are crucial tools in human genetic research, providing known landmarks for locating disease genes and frameworks for large-scale sequencing.Radiation hybrid mapping is one technique for building genome maps.In this paper, we describe the methods used to build radiation hybrid maps of the entire human genome.We present the hidden Markov model that we employ to estimate the likelihood of a map despite uncertainty about the data, and we discuss the problem of searching for maximum-likelihood maps.We describe the graph algorithms used to find sparse but reliable init,ial maps and our methods of extending them.Finally, we show results validating our software on simulated data, and we describe our genome-wide human radiat,ion hybrid maps and the evidence supporting them.

We introduce a new model for learning with membership queries in which queries near the boundary of a target concept may receive incorrect or "don't care" responses. In partial compensation, we assume the distribution of examples has zero probability mass on the boundary region. The motivation behind this model is that the reason for the incorrect (or "don't care") response is that these examples are extremely rare in practice. Thus, it does not matter how the learner classifies them. We present several positive results in this new model. We show how to learn the intersection of two halfspaces when membership queries near the boundary may be answered incorrectly. Our algorithm is an extension of an algorithmof Baum [6, 5] which learns intersections of two homogeneous halfspaces in the PAC-with-membership-queries model. We also describe algorithms for learning several subclasses of monotone DNF formulas. 1 Introduction In most of the theoretical work in concept learning it is assumed t...

We show that two cooperating robots can learn exactly any strongly-connected directed graph with n indistinguishable nodes in expected time polynomial in n. We introduce a new type of homing sequence for two robots which helps the robots recognize certain previously-seen nodes. We then present an algorithm in which the robots learn the graph and the homing sequence simultaneously by wandering actively through the graph. Unlike most previous learning results using homing sequences, our algorithm does not require a teacher to provide counterexamples. Furthermore, the algorithm can use efficiently any additional information available that distinguishes nodes. We also present an algorithm in which the robots learn by taking random walks. The rate at which a random walk converges to the stationary distribution is characterized by the conductance of the graph. Our random-walk algorithm learns in expected time polynomial in n and in the inverse of the conductance and is more efficient than the homing-sequence algorithm for high-conductance graphs.>