Prioritization of candidate genes for attention deficit hyperactivity disorder by computational analysis of multiple data sources

Attention deficit hyperactivity disorder(ADHD)is a common,highly heritable psychiatric disorder charac-terized by hyperactivity,inattention and increased im-pulsivity.In recent years,a large number of genetic studies for ADHD have been published and related ge-netic data has been accumulated dramatically.To pro-vide researchers a comprehensive ADHD genetic re-source,we previously developed the first genetic data-base for ADHD(ADHDgene).The abundant genetic data provides novel candidates for further study.Meanwhile,it also brings new challenge for selecting promising candidate genes for replication and verification research.In this study,we surveyed the computational tools for candidate gene prioritization and selected five tools,which integrate multiple data sources for gene prioritiza-tion,to prioritize ADHD candidate genes in ADHDgene.The prioritization analysis resulted in 16 prioritized can-didate genes,which are mainly involved in several major neurotransmitter systems or in nervous system development pathways.Among these genes,nervous system development related genes,especially SNAP25,STX1A and the gene-gene interactions related with each of them deserve further investigations.Our results may provide new insight for further verification study and facilitate the exploration of pathogenesis mechanism of ADHD.

Predicting Stock Price Movement with Multiple Data Sources and Machine Learning Models

Stock price trend prediction is a challenging issue in the financial field.To get improvements in predictive performance,both data and technique are essential.The purpose of this paper is to compare deep learning model(LSTM)with two ensemble models(RF and XGboost)using multiple data.Data is gathered from four stocks of financial sector in China A-share market,and the accuracy and F1-measure are used as performance measure.The data of the past three days is applied to classify the rise and fall trend of price on the next day.The models’performance are tested under different market styles(bull or bear market)and different market activities.The results indicate that under the same conditions,LSTM is the top algorithm followed by RF and XGBoost.For all models applied in this study,prediction performance in bull markets is much better than in bear markets,and the result in active period is better than inactive period by average.It is also found that adding data sources is not always effective in improving forecasting performance,and valuable data sources and proper processing may be more essential than providing a large quantity of data source.