Abstract
There is a growing interest in applying machine learning algorithms to real-world examples by explicitly deriving models based on probabilistic reasoning. Sports analytics, being favoured ...mostly by the statistics community and less discussed in the machine learning community, becomes our focus in this paper. Specifically, we model two-team sports for the sake of one-match-ahead forecasting. We present a pioneering modeling approach based on stacked Bayesian regressions, in a way that winning probability can be calculated analytically. Benefiting from regression flexibility and high standard of performance, Sparse Spectrum Gaussian Process Regression (SSGPR) – an improved algorithm for the standard Gaussian Process Regression (GPR), was used to solve Bayesian regression tasks, resulting in a novel predictive model called TLGProb. For evaluation, TLGProb was applied to a popular sports event – National Basketball Association (NBA). Finally, 85.28% of the matches in NBA 2014/2015 regular season were correctly predicted by TLGProb, surpassing the existing predictive models for NBA.
Abstract
Summary
Genome-wide association study (GWAS) analyses, at sufficient sample sizes and power, have successfully revealed biological insights for several complex traits. RICOPILI, an ...open-sourced Perl-based pipeline was developed to address the challenges of rapidly processing large-scale multi-cohort GWAS studies including quality control (QC), imputation and downstream analyses. The pipeline is computationally efficient with portability to a wide range of high-performance computing environments. RICOPILI was created as the Psychiatric Genomics Consortium pipeline for GWAS and adopted by other users. The pipeline features (i) technical and genomic QC in case-control and trio cohorts, (ii) genome-wide phasing and imputation, (iv) association analysis, (v) meta-analysis, (vi) polygenic risk scoring and (vii) replication analysis. Notably, a major differentiator from other GWAS pipelines, RICOPILI leverages on automated parallelization and cluster job management approaches for rapid production of imputed genome-wide data. A comprehensive meta-analysis of simulated GWAS data has been incorporated demonstrating each step of the pipeline. This includes all the associated visualization plots, to allow ease of data interpretation and manuscript preparation. Simulated GWAS datasets are also packaged with the pipeline for user training tutorials and developer work.
Availability and implementation
RICOPILI has a flexible architecture to allow for ongoing development and incorporation of newer available algorithms and is adaptable to various HPC environments (QSUB, BSUB, SLURM and others). Specific links for genomic resources are either directly provided in this paper or via tutorials and external links. The central location hosting scripts and tutorials is found at this URL: https://sites.google.com/a/broadinstitute.org/RICOPILI/home
Supplementary information
Supplementary data are available at Bioinformatics online.
One of the leading single-channel speech separation (SS) models is based on a TasNet with a dual-path segmentation technique, where the size of each segment remains unchanged throughout all layers. ...In contrast, our key finding is that multi-granularity features are essential for enhancing contextual modeling and computational efficiency. We introduce a self-attentive network with a novel sandglass-shape, namely Sandglasset, which advances the state-of-the-art (SOTA) SS performance at significantly smaller model size and computational cost. Forward along each block inside Sandglasset, the temporal granularity of the features gradually becomes coarser until reaching half of the network blocks, and then successively turns finer towards the raw signal level. We also unfold that residual connections between features with the same granularity are critical for preserving information after passing through the bottleneck layer. Experiments show our Sandglasset with only 2.3M parameters has achieved the best results on two benchmark SS datasets - WSJ0-2mix and WSJ0-3mix, where the SI-SNRi scores have been improved by absolute 0.6 dB and 2.4 dB, respectively, comparing to the prior SOTA results.
Deep-learning based speech separation models confront poor generalization problem that even the state-of-the-art models could abruptly fail when evaluating them in mismatch conditions. To address ...this problem, we propose an easy-to-implement yet effective consistency based semi-supervised learning (SSL) approach, namely Mixup-Breakdown training (MBT). It learns a teacher model to "breakdown" unlabeled inputs, and the estimated separations are interpolated to produce more useful pseudo "mixup" input-output pairs, on which the consistency regularization could apply for learning a student model. In our experiment, we evaluate MBT under various conditions with ascending degrees of mismatch, including unseen interfering speech, noise, and music, and compare MBT's generalization capability against state-of-the-art supervised learning and SSL approaches. The result indicates that MBT significantly outperforms several strong baselines with up to 13.77% relative SI-SNRi improvement. Moreover, MBT only adds negligible computational overhead to standard training schemes.
To extract robust deep representations from long sequential modeling of speech data, we propose a self-supervised learning approach, namely Contrastive Separative Coding (CSC). Our key finding is to ...learn such representations by separating the target signal from contrastive interfering signals. First, a multi-task separative encoder is built to extract shared separable and discriminative embedding; secondly, we propose a powerful cross-attention mechanism performed over speaker representations across various interfering conditions, allowing the model to focus on and globally aggregate the most critical information to answer the "query" (current bottom-up embedding) while paying less attention to interfering, noisy, or irrelevant parts; lastly, we form a new probabilistic contrastive loss which estimates and maximizes the mutual information between the representations and the global speaker vector. While most prior unsupervised methods have focused on predicting the future, neighboring, or missing samples, we take a different perspective of predicting the interfered samples. Moreover, our contrastive separative loss is free from negative sampling. The experiment demonstrates that our approach can learn useful representations achieving a strong speaker verification performance in adverse conditions.
IMPORTANCE: Schizophrenia is associated with widespread cognitive impairments. Although cognitive deficits are one of the factors most strongly associated with functional outcome in schizophrenia, ...current treatment strategies largely fail to ameliorate these impairments. To develop more efficient treatment strategies in patients with schizophrenia, a better understanding of the pathogenesis of these cognitive deficits is needed. Accumulating evidence indicates that genetic risk of schizophrenia may contribute to cognitive dysfunction. OBJECTIVE: To identify genomic regions jointly influencing schizophrenia and the cognitive domains of reaction time and verbal-numerical reasoning, as well as general cognitive function, a phenotype that captures the shared variation in performance across cognitive domains. DESIGN, SETTING, AND PARTICIPANTS: Combining data from genome-wide association studies from multiple phenotypes using conditional false discovery rate analysis provides increased power to discover genetic variants and could elucidate shared molecular genetic mechanisms. Data from the following genome-wide association studies, published from July 24, 2014, to January 17, 2017, were combined: schizophrenia in the Psychiatric Genomics Consortium cohort (n = 79 757 cases, 34 486; controls, 45 271); verbal-numerical reasoning (n = 36 035) and reaction time (n = 111 483) in the UK Biobank cohort; and general cognitive function in CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology) (n = 53 949) and COGENT (Cognitive Genomics Consortium) (n = 27 888). MAIN OUTCOMES AND MEASURES: Genetic loci identified by conditional false discovery rate analysis. Brain messenger RNA expression and brain expression quantitative trait locus functionality were determined. RESULTS: Among the participants in the genome-wide association studies, 21 loci jointly influencing schizophrenia and cognitive traits were identified: 2 loci shared between schizophrenia and verbal-numerical reasoning, 6 loci shared between schizophrenia and reaction time, and 14 loci shared between schizophrenia and general cognitive function. One locus was shared between schizophrenia and 2 cognitive traits and represented the strongest shared signal detected (nearest gene TCF20; chromosome 22q13.3), and was shared between schizophrenia (z score, 5.01; P = 5.53 × 10−7), general cognitive function (z score, –4.43; P = 9.42 × 10−6), and verbal-numerical reasoning (z score, –5.43; P = 5.64 × 10−8). For 18 loci, schizophrenia risk alleles were associated with poorer cognitive performance. The implicated genes are expressed in the developmental and adult human brain. Replicable expression quantitative trait locus functionality was identified for 4 loci in the adult human brain. CONCLUSIONS AND RELEVANCE: The discovered loci improve the understanding of the common genetic basis underlying schizophrenia and cognitive function, suggesting novel molecular genetic mechanisms.
Individuals at ultra-high risk (UHR) of psychosis are characterised by the emergence of attenuated psychotic symptoms and deterioration in functioning. In view of the high non-psychotic comorbidity ...and low rates of transition to psychosis, the specificity of the UHR status has been called into question. This study aims to (i) investigate if the UHR construct is associated with the genetic liability of schizophrenia or other psychiatric conditions; (ii) examine the ability of polygenic risk scores (PRS) to discriminate healthy controls from UHR, remission and conversion status. PRS was calculated for 210 youths (nUHR = 102, nControl = 108) recruited as part of the Longitudinal Youth at Risk Study (LYRIKS) using nine psychiatric traits derived from twelve large-scale psychiatric genome-wide association studies as discovery datasets. PRS was also examined to discriminate UHR-Healthy control status, and healthy controls from UHR remission and conversion status. Result indicated that schizophrenia PRS appears to best index the genetic liability of UHR, while trend level associations were observed for depression and cross-disorder PRS. Schizophrenia PRS discriminated healthy controls from UHR (R2 = 7.9%, p = 2.59 x 10-3, OR = 1.82), healthy controls from non-remitters (R2 = 8.1%, p = 4.90 x 10-4, OR = 1.90), and converters (R2 = 7.6%, p = 1.61 x 10-3, OR = 1.82), with modest predictive ability. A trend gradient increase in schizophrenia PRS was observed across categories. The association between schizophrenia PRS and UHR status supports the hypothesis that the schizophrenia polygenic liability indexes the risk for developing psychosis.
Abstract Recent studies have reported a high prevalence of psychiatric comorbidities in Ultra High Risk (UHR) for psychosis populations. This study examined the prevalence of comorbidity and its ...impact on symptoms, functioning, cognition and transition to psychosis in the Longitudinal Youth at Risk Study (LYRIKS) sample. The Comprehensive Assessment of At-Risk Mental State (CAARMS) was used to identify UHR individuals and 163 participants were included in the study. Comorbid disorders were identified using the Structured Clinical Interview for DSM-IV-TR Axis I Disorders. Participants were evaluated on the CAARMS, Positive and Negative Syndrome Scale, Calgary Depression Scale for Schizophrenia, Beck Anxiety Inventory, Global Assessment of Functioning and Brief Assessment of Cognition in Schizophrenia. Clinical, functioning and cognitive characteristics by lifetime and current comorbidity groups were compared using multivariate tests. Independent predictors of comorbidity were identified through logistic regression. Chi-squared tests were used to compare comorbidity rates between those who had developed psychosis at one year and those who had not. We found that 131 UHR participants (80.4%) had a lifetime comorbidity while 82 (50.3%) had a current comorbidity with depressive disorders being the most common. UHR individuals with comorbidity had more severe symptoms, higher distress and lower functioning with no differences in general cognition. Lower functioning was associated with current comorbidity. Eleven participants (6.7%) had developed psychosis after one year and there were no differences in the comorbidity rates between those who developed psychosis and those who did not. Psychiatric comorbidities in the UHR group are associated with adverse clinical outcomes and warrant closer attention.
Earlier studies examining structural brain abnormalities associated with cognitively derived subgroups were mainly cross-sectional in design and had mixed findings. Thus, we obtained cross-sectional ...and longitudinal data to characterize the extent and trajectory of brain structure abnormalities underlying distinct cognitive subtypes ("preserved," "deteriorated," and "compromised") seen in psychotic spectrum disorders.
Data from 364 subjects (225 patients with psychotic conditions and 139 healthy controls) were first used to determine the relationship of cognitive subtypes with cross-sectional measures of subcortical volume and cortical thickness. To probe neurodevelopmental abnormalities, brain structure laterality was examined. To examine whether neuroprogressive abnormalities persist, longitudinal brain structural changes over 5 years were examined within a subset of 101 subjects. Subsequent discriminant analysis using the identified brain measures was performed on an independent subject group.
Cross-sectional comparisons showed that cortical thinning and limbic volume reductions were most widespread in "deteriorated" cognitive subtype. Laterality comparisons showed more rightward amygdala lateralization in "compromised" than "preserved" subtype. Longitudinal comparisons revealed progressive hippocampal shrinkage in "deteriorated" compared with healthy controls and "preserved" subtype, which correlated with worse negative symptoms, cognitive and psychosocial functioning. Post-hoc discrimination analysis on an independent group of 52 subjects using the identified brain structures found an overall accuracy of 71% for classification of cognitive subtypes.
These findings point toward distinct extent and trajectory of corticolimbic abnormalities associated with cognitive subtypes in psychosis, which can allow further understanding of the biological course of cognitive functioning over illness course and with treatment.
Schizophrenia is characterized by neuropsychological deficits across many cognitive domains. Cognitive phenotypes with high heritability and genetic overlap with schizophrenia liability can help ...elucidate the mechanisms leading from genes to psychopathology. We performed a meta-analysis of 170 published twin and family heritability studies of >800 000 nonpsychiatric and schizophrenia subjects to accurately estimate heritability across many neuropsychological tests and cognitive domains. The proportion of total variance of each phenotype due to additive genetic effects (A), shared environment (C), and unshared environment and error (E), was calculated by averaging A, C, and E estimates across studies and weighting by sample size. Heritability ranged across phenotypes, likely due to differences in genetic and environmental effects, with the highest heritability for General Cognitive Ability (32%-67%), Verbal Ability (43%-72%), Visuospatial Ability (20%-80%), and Attention/Processing Speed (28%-74%), while the lowest heritability was observed for Executive Function (20%-40%). These results confirm that many cognitive phenotypes are under strong genetic influences. Heritability estimates were comparable in nonpsychiatric and schizophrenia samples, suggesting that environmental factors and illness-related moderators (eg, medication) do not substantially decrease heritability in schizophrenia samples, and that genetic studies in schizophrenia samples are informative for elucidating the genetic basis of cognitive deficits. Substantial genetic overlap between cognitive phenotypes and schizophrenia liability (average rg = -.58) in twin studies supports partially shared genetic etiology. It will be important to conduct comparative studies in well-powered samples to determine whether the same or different genes and genetic variants influence cognition in schizophrenia patients and the general population.