TrackUSF, a novel tool for automated ultrasonic vocalization analysis, reveals modified calls in a rat model of autism.

BACKGROUND: Various mammalian species emit ultrasonic vocalizations (USVs), which reflect their emotional state and mediate social interactions. USVs are usually analyzed by manual or semi-automated methodologies that categorize discrete USVs according to their structure in the frequency-time domains. This laborious analysis hinders the effective use of USVs as a readout for high-throughput analysis of behavioral changes in animals. RESULTS: Here we present a novel automated open-source tool that utilizes a different approach towards USV analysis, termed TrackUSF. To validate TrackUSF, we analyzed calls from different animal species, namely mice, rats, and bats, recorded in various settings and compared the results with a manual analysis by a trained observer. We found that TrackUSF detected the majority of USVs, with less than 1% of false-positive detections. We then employed TrackUSF to analyze social vocalizations in Shank3-deficient rats, a rat model of autism, and revealed that these vocalizations exhibit a spectrum of deviations from appetitive calls towards aversive calls. CONCLUSIONS: TrackUSF is a simple and easy-to-use system that may be used for a high-throughput comparison of ultrasonic vocalizations between groups of animals of any kind in any setting, with no prior assumptions.

Association of Social Distancing, Population Density, and Temperature With the Instantaneous Reproduction Number of SARS-CoV-2 in Counties Across the United States.

Importance: Local variation in the transmission of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) across the United States has not been well studied. Objective: To examine the association of county-level factors with variation in the SARS-CoV-2 reproduction number over time. Design, Setting, and Participants: This cohort study included 211 counties, representing state capitals and cities with at least 100 000 residents and including 178 892 208 US residents, in 46 states and the District of Columbia between February 25, 2020, and April 23, 2020. Exposures: Social distancing, measured by percentage change in visits to nonessential businesses; population density; and daily wet-bulb temperatures. Main Outcomes and Measures: Instantaneous reproduction number (Rt), or cases generated by each incident case at a given time, estimated from daily case incidence data. Results: The 211 counties contained 178 892 208 of 326 289 971 US residents (54.8%). Median (interquartile range) population density was 1022.7 (471.2-1846.0) people per square mile. The mean (SD) peak reduction in visits to nonessential business between April 6 and April 19, as the country was sheltering in place, was 68.7% (7.9%). Median (interquartile range) daily wet-bulb temperatures were 7.5 (3.8-12.8) °C. Median (interquartile range) case incidence and fatality rates per 100 000 people were approximately 10 times higher for the top decile of densely populated counties (1185.2 [313.2-1891.2] cases; 43.7 [10.4-106.7] deaths) than for counties in the lowest density quartile (121.4 [87.8-175.4] cases; 4.2 [1.9-8.0] deaths). Mean (SD) Rt in the first 2 weeks was 5.7 (2.5) in the top decile compared with 3.1 (1.2) in the lowest quartile. In multivariable analysis, a 50% decrease in visits to nonessential businesses was associated with a 45% decrease in Rt (95% CI, 43%-49%). From a relative Rt at 0 °C of 2.13 (95% CI, 1.89-2.40), relative Rt decreased to a minimum as temperatures warmed to 11 °C, increased between 11 and 20 °C (1.61; 95% CI, 1.42-1.84) and then declined again at temperatures greater than 20 °C. With a 70% reduction in visits to nonessential business, 202 counties (95.7%) were estimated to fall below a threshold Rt of 1.0, including 17 of 21 counties (81.0%) in the top density decile and 52 of 53 counties (98.1%) in the lowest density quartile.2. Conclusions and Relevance: In this cohort study, social distancing, lower population density, and temperate weather were associated with a decreased Rt for SARS-CoV-2 in counties across the United States. These associations could inform selective public policy planning in communities during the coronavirus disease 2019 pandemic.

Learn on Source, Refine on Target: A Model Transfer Learning Framework with Random Forests.

We propose novel model transfer-learning methods that refine a decision forest model M learned within a "source" domain using a training set sampled from a "target" domain, assumed to be a variation of the source. We present two random forest transfer algorithms. The first algorithm searches greedily for locally optimal modifications of each tree structure by trying to locally expand or reduce the tree around individual nodes. The second algorithm does not modify structure, but only the parameter (thresholds) associated with decision nodes. We also propose to combine both methods by considering an ensemble that contains the union of the two forests. The proposed methods exhibit impressive experimental results over a range of problems.

Automated gene-model curation using global discriminative learning.

MOTIVATION: Gene-model curation creates consensus gene models by combining multiple sources of protein-coding evidence that may be incomplete or inconsistent. To date, manual curation still produces the highest quality models. However, manual curation is too slow and costly to be completed even for the most important organisms. In recent years, machine-learned ensemble gene predictors have become a viable alternative to manual curation. Current approaches make use of signal and genomic region consistency among sources and some voting scheme to resolve conflicts in the evidence. As a further step in that direction, we have developed eCRAIG (ensemble CRAIG), an automated curation tool that combines multiple sources of evidence using global discriminative training. This allows efficient integration of different types of genomic evidence with complex statistical dependencies to maximize directly annotation accuracy. Our method goes beyond previous work in integrating novel non-linear annotation agreement features, as well as combinations of intrinsic features of the target sequence and extrinsic annotation features. RESULTS: We achieved significant improvements over the best ensemble predictors available for Homo sapiens, Caenorhabditis elegans and Arabidopsis thaliana. In particular, eCRAIG achieved a relative mean improvement of 5.1% over Jigsaw, the best published ensemble predictor in all our experiments. AVAILABILITY: The source code and datasets are both available at http://www.seas.upenn.edu/abernal/ecraig.tgz.

Reranking candidate gene models with cross-species comparison for improved gene prediction.

BACKGROUND: Most gene finders score candidate gene models with state-based methods, typically HMMs, by combining local properties (coding potential, splice donor and acceptor patterns, etc). Competing models with similar state-based scores may be distinguishable with additional information. In particular, functional and comparative genomics datasets may help to select among competing models of comparable probability by exploiting features likely to be associated with the correct gene models, such as conserved exon/intron structure or protein sequence features. RESULTS: We have investigated the utility of a simple post-processing step for selecting among a set of alternative gene models, using global scoring rules to rerank competing models for more accurate prediction. For each gene locus, we first generate the K best candidate gene models using the gene finder Evigan, and then rerank these models using comparisons with putative orthologous genes from closely-related species. Candidate gene models with lower scores in the original gene finder may be selected if they exhibit strong similarity to probable orthologs in coding sequence, splice site location, or signal peptide occurrence. Experiments on Drosophila melanogaster demonstrate that reranking based on cross-species comparison outperforms the best gene models identified by Evigan alone, and also outperforms the comparative gene finders GeneWise and Augustus+. CONCLUSION: Reranking gene models with cross-species comparison improves gene prediction accuracy. This straightforward method can be readily adapted to incorporate additional lines of evidence, as it requires only a ranked source of candidate gene models.

Global discriminative learning for higher-accuracy computational gene prediction.

Most ab initio gene predictors use a probabilistic sequence model, typically a hidden Markov model, to combine separately trained models of genomic signals and content. By combining separate models of relevant genomic features, such gene predictors can exploit small training sets and incomplete annotations, and can be trained fairly efficiently. However, that type of piecewise training does not optimize prediction accuracy and has difficulty in accounting for statistical dependencies among different parts of the gene model. With genomic information being created at an ever-increasing rate, it is worth investigating alternative approaches in which many different types of genomic evidence, with complex statistical dependencies, can be integrated by discriminative learning to maximize annotation accuracy. Among discriminative learning methods, large-margin classifiers have become prominent because of the success of support vector machines (SVM) in many classification tasks. We describe CRAIG, a new program for ab initio gene prediction based on a conditional random field model with semi-Markov structure that is trained with an online large-margin algorithm related to multiclass SVMs. Our experiments on benchmark vertebrate datasets and on regions from the ENCODE project show significant improvements in prediction accuracy over published gene predictors that use intrinsic features only, particularly at the gene level and on genes with long introns.

Online ranking by projecting.

We discuss the problem of ranking instances. In our framework, each instance is associated with a rank or a rating, which is an integer in 1 to k. Our goal is to find a rank-prediction rule that assigns each instance a rank that is as close as possible to the instance's true rank. We discuss a group of closely related online algorithms, analyze their performance in the mistake-bound model, and prove their correctness. We describe two sets of experiments, with synthetic data and with the EachMovie data set for collaborative filtering. In the experiments we performed, our algorithms outperform online algorithms for regression and classification applied to ranking.