Benchmarking deep learning methods for predicting CRISPR/Cas9 sgRNA on- and off-target activities.

In silico design of single guide RNA (sgRNA) plays a critical role in clustered regularly interspaced, short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) system. Continuous efforts are aimed at improving sgRNA design with efficient on-target activity and reduced off-target mutations. In the last 5 years, an increasing number of deep learning-based methods have achieved breakthrough performance in predicting sgRNA on- and off-target activities. Nevertheless, it is worthwhile to systematically evaluate these methods for their predictive abilities. In this review, we conducted a systematic survey on the progress in prediction of on- and off-target editing. We investigated the performances of 10 mainstream deep learning-based on-target predictors using nine public datasets with different sample sizes. We found that in most scenarios, these methods showed superior predictive power on large- and medium-scale datasets than on small-scale datasets. In addition, we performed unbiased experiments to provide in-depth comparison of eight representative approaches for off-target prediction on 12 publicly available datasets with various imbalanced ratios of positive/negative samples. Most methods showed excellent performance on balanced datasets but have much room for improvement on moderate- and severe-imbalanced datasets. This study provides comprehensive perspectives on CRISPR/Cas9 sgRNA on- and off-target activity prediction and improvement for method development.

Deep learning-based recurrence detector on magnetic resonance scans in nasopharyngeal carcinoma: A multicenter study.

OBJECTIVES: Accuracy in the detection of recurrent nasopharyngeal carcinoma (NPC) on follow-up magnetic resonance (MR) scans needs to be improved. MATERIAL AND METHODS: A total of 5 035 follow-up MR scans from 5 035 survivors with treated NPC between April 2007 and July 2020 were retrospectively collected from three cancer centers for developing and evaluating the deep learning (DL) model MODERN (MR-based Deep learning model for dEtecting Recurrent Nasopharyngeal carcinoma). In a reader study with 220 scans, the accuracy of two radiologists in detecting recurrence on scans with vs without MODERN was evaluated. The performance was measured using the area under the receiver operating characteristic curve (ROC-AUC) and accuracy with a 95% confidence interval (CI). RESULTS: MODERN exhibited sound performance in the validation cohort (internal: ROC-AUC, 0.88, 95% CI, 0.86-0.90; external 1: ROC-AUC, 0.88, 95% CI, 0.86-0.90; external 2: ROC-AUC, 0.85, 95% CI, 0.82-0.88). In a reader study, MODERN alone achieved reliable accuracy compared to that of radiologists (MODERN: 84.1%, 95% CI, 79.3%-88.9%; competent: 80.9%, 95% CI, 75.7%-86.1%, P < 0.001; expert: 85.9%, 95% CI, 81.3%-90.5%, P < 0.001). The accuracy of radiologists was boosted by the MODERN score (competent with MODERN score: 84.6%, 95% CI, 79.8%-89.3%, P < 0.001; expert with MODERN score: 87.7%, 95% CI, 83.4%-92.1%, P < 0.001). CONCLUSION: We developed a DL model for recurrence detection with reliable performance. Computer-human collaboration has the potential to refine the workflow in interpreting surveillant MR scans among patients with treated NPC.

The mechanism accounting for DNA damage strength modulation of p53 dynamical properties.

The P53 protein levels exhibit a series of pulses in response to DNA double-stranded breaks (DSBs). However, the mechanism regarding how damage strength regulates physical parameters of p53 pulses remains to be elucidated. This paper established two mathematical models translating the mechanism of p53 dynamics in response to DSBs; the two models can reproduce many results observed in the experiments. Based on the models, numerical analysis suggested that the interval between pulses increases as the damage strength decreases, and we proposed that the p53 dynamical system in response to DSBs is modulated by frequency. Next, we found that the ATM positive self-feedback can realize the system characteristic that the pulse amplitude is independent of the damage strength. In addition, the pulse interval is negatively correlated with apoptosis; the greater the damage strength, the smaller the pulse interval, the faster the p53 accumulation rate, and the cells are more susceptible to apoptosis. These findings advance our understanding of the mechanism of p53 dynamical response and give new insights for experiments to probe the dynamics of p53 signaling.

The Influence of Culture Capital, Social Security, and Living Conditions on Children's Cognitive Ability: Evidence from 2018 China Family Panel Studies.

The aim of this study was to analyze the influence of economic capital, culture capital, social capital, social security, and living conditions on children's cognitive ability. However, most studies only focus on the impact of family socio-economic status/culture capital on children's cognitive ability by ordinary least squares regression analysis. To this end, we used the data from the China Family Panel Studies in 2018 and applied proxy variable, instrumental variables, and two-stage least squares regression analysis with a total of 2647 samples with ages from 6 to 16. The results showed that family education, education expectation, books, education participation, social communication, and tap water had a positive impact on both the Chinese and math cognitive ability of children, while children's age, gender, and family size had a negative impact on cognitive ability, and the impact of genes was attenuated by family capital. In addition, these results are robust, and the heterogeneity was found for gender and urban location. Specifically, in terms of gender, the culture, social capital, and social security are more sensitive to the cognitive ability of girls, while living conditions are more sensitive to the cognitive ability of boys. In urban locations, the culture and social capital are more sensitive to rural children's cognitive ability, while the social security and living conditions are more sensitive to urban children's cognitive ability. These findings provide theoretical support to further narrow the cognitive differences between children from many aspects, which allows social security and living conditions to be valued.

SAAED: Embedding and Deep Learning Enhance Accurate Prediction of Association Between circRNA and Disease.

Emerging evidence indicates that circRNA can regulate various diseases. However, the mechanisms of circRNA in these diseases have not been fully understood. Therefore, detecting potential circRNA-disease associations has far-reaching significance for pathological development and treatment of these diseases. In recent years, deep learning models are used in association analysis of circRNA-disease, but a lack of circRNA-disease association data limits further improvement. Therefore, there is an urgent need to mine more semantic information from data. In this paper, we propose a novel method called Semantic Association Analysis by Embedding and Deep learning (SAAED), which consists of two parts, a neural network embedding model called Entity Relation Network (ERN) and a Pseudo-Siamese network (PSN) for analysis. ERN can fuse multiple sources of data and express the information with low-dimensional embedding vectors. PSN can extract the feature between circRNA and disease for the association analysis. CircRNA-disease, circRNA-miRNA, disease-gene, disease-miRNA, disease-lncRNA, and disease-drug association information are used in this paper. More association data can be introduced for analysis without restriction. Based on the CircR2Disease benchmark dataset for evaluation, a fivefold cross-validation experiment showed an AUC of 98.92%, an accuracy of 95.39%, and a sensitivity of 93.06%. Compared with other state-of-the-art models, SAAED achieves the best overall performance. SAAED can expand the expression of the biological related information and is an efficient method for predicting potential circRNA-disease association.

Generalized Poisson Hurdle Model for Count Data and Its Application in Ear Disease.

For count data, though a zero-inflated model can work perfectly well with an excess of zeroes and the generalized Poisson model can tackle over- or under-dispersion, most models cannot simultaneously deal with both zero-inflated or zero-deflated data and over- or under-dispersion. Ear diseases are important in healthcare, and falls into this kind of count data. This paper introduces a generalized Poisson Hurdle model that work with count data of both too many/few zeroes and a sample variance not equal to the mean. To estimate parameters, we use the generalized method of moments. In addition, the asymptotic normality and efficiency of these estimators are established. Moreover, this model is applied to ear disease using data gained from the New South Wales Health Research Council in 1990. This model performs better than both the generalized Poisson model and the Hurdle model.

Impact of Education, Medical Services, and Living Conditions on Health: Evidence from China Health and Nutrition Survey.

Education, medical services, and living conditions can influence individual health and health literacy. We used the 2015 China Health and Nutrition Survey data to analyze the impact of education, medical services, and living conditions on individual health by extending the Grossman model. As a result, using the instrumental variable (read, write, and draw) two-stage least square method, we found that education, medical services, and living conditions have a positive impact on individual health, both physical health and psychological health. Among them, medical services have the largest influence, followed by living conditions and education. In addition, the results are robust. However, individual characteristics, family income, and working status also affect individual health. Moreover, we observed heterogeneity in age, sex, and residence in the impact of education, medical services, and living conditions on individual health. In particular, the health of the rural elderly and elderly women is more sensitive to education, the medical services of middle-aged women and young men have a greater impact on their health, and the living conditions of the rural elderly and youth have a greater impact on their health. All the findings are helpful for optimizing the path of the Healthy China program.

Effect of family "upward" intergenerational support on the health of rural elderly in China: Evidence from Chinese Longitudinal Healthy Longevity Survey.

As health challenging rural elderly in an aging population, more attention is being paid on impact of family intergenerational support on the health of the elderly. This paper investigates the effects of children's intergenerational economic support and non-economic support on physical, mental, and functional health of rural elderly in China in the mean while. This paper applies the 2014 Chinese Longitudinal Healthy Longevity Survey (CLHLS), in particular, applying exploratory factor analysis to ascertain latent variables and Structural Equation Model (SEM), and analyzes the impacts of "Upward" intergenerational support on health of rural elderly. As resulted, after controlling the socioeconomic status of the rural elderly, the family "upward" intergenerational support influences the elderly's physical health at a percentage of 11.7%, mental health 29.8%, and physiological function 12.6%. Moreover, "Upward" economic support has a positive effect on physiological function (P<0.05). "Upward" non-economic support has negative effects on physiological function and mental health (P<0.05), while it has a positive effect on physical health. In addition, economically independent rural elderly are more likely to benefit from the health of "upward" intergenerational support, especially mental health. In particular, those results are robust. "Upward" intergenerational support plays an important role for the health of rural elderly. For the rural elderly of economic independence, to improve the quality of care and spiritual support, it is important to solve the health problems. In addition, it is necessary to build a comprehensive old-age security and support system for family, community, and society jointly to improve the health of the rural elderly.

Prediction of CRISPR/Cas9 single guide RNA cleavage efficiency and specificity by attention-based convolutional neural networks.

CRISPR/Cas9 is a preferred genome editing tool and has been widely adapted to ranges of disciplines, from molecular biology to gene therapy. A key prerequisite for the success of CRISPR/Cas9 is its capacity to distinguish between single guide RNAs (sgRNAs) on target and homologous off-target sites. Thus, optimized design of sgRNAs by maximizing their on-target activity and minimizing their potential off-target mutations are crucial concerns for this system. Several deep learning models have been developed for comprehensive understanding of sgRNA cleavage efficacy and specificity. Although the proposed methods yield the performance results by automatically learning a suitable representation from the input data, there is still room for the improvement of accuracy and interpretability. Here, we propose novel interpretable attention-based convolutional neural networks, namely CRISPR-ONT and CRISPR-OFFT, for the prediction of CRISPR/Cas9 sgRNA on- and off-target activities, respectively. Experimental tests on public datasets demonstrate that our models significantly yield satisfactory results in terms of accuracy and interpretability. Our findings contribute to the understanding of how RNA-guide Cas9 nucleases scan the mammalian genome. Data and source codes are available at https://github.com/Peppags/CRISPRont-CRISPRofft.

The Impact of Social Capital on Mental Health: Evidence from the China Family Panel Survey.

The influence of social capital on mental health is a controversial topic. As some studies have pointed out, cognitive social capital significantly affects mental health but structural social capital does not. Using data from the China Family Panel Survey, this study measured social capital from social help, social trust, social networks, and social participation, and took regional average level of social capital as the instrumental variables, and applied a two-stage least squares regression. We found that the mental health of residents who trust and help each other is significantly higher than that of residents without trust and mutual help. When residents' efforts to maintain social networks increase, their mental health significantly improves. These results are robust. Furthermore, the impact of social capital on mental health was heterogeneous in terms of urbanicity, gender, age, and area. These results are helpful for making policies for promoting residents' mental health.

Identifying Circular RNA and Predicting Its Regulatory Interactions by Machine Learning.

Circular RNA (circRNA) is a closed long non-coding RNA (lncRNA) formed by covalently closed loops through back-splicing. Emerging evidence indicates that circRNA can influence cellular physiology through various molecular mechanisms. Thus, accurate circRNA identification and prediction of its regulatory information are critical for understanding its biogenesis. Although several computational tools based on machine learning have been proposed for circRNA identification, the prediction accuracy remains to be improved. Here, first we present circLGB, a machine learning-based framework to discriminate circRNA from other lncRNAs. circLGB integrates commonly used sequence-derived features and three new features containing adenosine to inosine (A-to-I) deamination, A-to-I density and the internal ribosome entry site. circLGB categorizes circRNAs by utilizing a LightGBM classifier with feature selection. Second, we introduce circMRT, an ensemble machine learning framework to systematically predict the regulatory information for circRNA, including their interactions with microRNA, the RNA binding protein, and transcriptional regulation. Feature sets including sequence-based features, graph features, genome context, and regulatory information features were modeled in circMRT. Experiments on public and our constructed datasets show that the proposed algorithms outperform the available state-of-the-art methods. circLGB is available at http://www.circlgb.com. Source codes are available at https://github.com/Peppags/circLGB-circMRT.

C-RNNCrispr: Prediction of CRISPR/Cas9 sgRNA activity using convolutional and recurrent neural networks.

CRISPR/Cas9 is a hot genomic editing tool, but its success is limited by the widely varying target efficiencies among different single guide RNAs (sgRNAs). In this study, we proposed C-RNNCrispr, a hybrid convolutional neural networks (CNNs) and bidirectional gate recurrent unit network (BGRU) framework, to predict CRISPR/Cas9 sgRNA on-target activity. C-RNNCrispr consists of two branches: sgRNA branch and epigenetic branch. The network receives the encoded binary matrix of sgRNA sequence and four epigenetic features as inputs, and produces a regression score. We introduced a transfer learning approach by using small-size datasets to fine-tune C-RNNCrispr model that were pre-trained from benchmark dataset, leading to substantially improved predictive performance. Experiments on commonly used datasets showed C-RNNCrispr outperforms the state-of-the-art methods in terms of prediction accuracy and generalization. Source codes are available at https://github.com/Peppags/C_RNNCrispr.

Exploring the Influence of Parameters on the p53 Response When Single-Stranded Breaks and Double-Stranded Breaks Coexist.

The p53 response to DNA damage is closely related to cell fate decisions. P53 preferentially responds to single-stranded breaks (SSBs) exhibiting a graded response when single-stranded breaks (SSBs) and double-stranded breaks (DSBs) coexist. However, how p53 natural preferential response is affected by kinetic parameters remains to be elucidated. Here, based on the hybrid model I, we computationally searched all the parameters and parameter combinations in the parameter space to identify those that could alter the natural preferential response of p53 when SSBs and DSBs coexist. Firstly, when a single parameter is changed, the parameters that can alter graded response to produce p53 pulse response are production rate of ATM- and Rad3-related kinase(ATR) (beta2), ATR degradation rate (alf2) and ATR-dependent p53 production rate (beta31). Secondly, when double parameters are changed, the combinations of beta2/alf2/beta31 and any other parameters are capable of altering the p53 natural preferential response, and the combination of ataxia-telangiectasia mutated kinase (ATM)-dependent p53 production rate (beta3) and Wip1-dependent p53 degradation rate (alf35) is also capable of altering the p53 natural preferential response. Thirdly, we analyzed the sensitivity of both pulse amplitude and apoptosis to kinetic parameters. We find that pulse amplitude is most sensitive to ATM-dependent p53 production rate (beta3), and apoptosis is more sensitive to damage-dependent ATM production rate (beta1), wip1-dependent ATM degradation rate (alf15), wip1 production rate (beta5) and wip1 degradation rate (alf5). What is more, the smaller the value of alf15/beta5 or the larger the value of beta1/alf5, the more susceptible the cells are to apoptosis. These results provide clues to design more effective and less toxic targeted treatments for cancer.

Insights into active intragenic enhancers.

Enhancers can regulate gene transcription from afar. Many enhancers are located in genes. Although the regulatory roles of several individual intragenic enhancers have been elaborated, a genome-wide insight into intragenic enhancers remains to be elucidated. We found that active intragenic enhancers have a preference for being located in expressed genes. Unlike intergenic enhancers, active intragenic enhancers are enriched of H3K79me2 epigenetic signal, and depleted of variant histone H2A.Z. Moreover, eRNAs of active intragenic enhancers show lower degradation rates than those of the other enhancers. Our findings will have implications in understanding functions of intragenic enhancers.

CCRDB: a cancer circRNAs-related database and its application in hepatocellular carcinoma-related circRNAs.

Circular RNAs (circRNAs) are widely expressed in human cells and tissues and can form a covalently closed exon circularization, which have stable patterns and play important regulatory roles in physiological or pathological process. There is still lack of a comprehensively disease-related knowledge base for in-depth analysis of circRNAs. In this paper, a cancer circRNAs-related database (CCRDB) was established. The CCRDB's initial circRNAs data were collected by sequencing experimental data of 10 samples from 5 patients with hepatocellular carcinoma (HCC), where a total of 11 501 circRNAs were found and can easily be expanded by collecting and analyzing external data sources such as circBASE (1). Using CCRDB, we have further studied the relationships between circRNAs and HCC and found that circRNAs (hsa_circ_ 0002130, hsa_circ_0084615, hsa_circ_0001445, hsa_circ_0001727 and hsa_circ_0001361) and the corresponding genes ID [C3 (2, 3), ASPH (4), SMARCA5 (5), ZKSCAN1 (6) and FNDC3B (7)], respectively, might be the potential biomarker targets for HCC. Furthermore, our experiment also found that some new circRNAs chromosome sites chr12:23998917 24048958 and chr16:72090429 72093087 and the corresponding genes ID (SOX5 (8) and HP (9), respectively), might be the potential biomarker targets for HCC. These results indicate that CCRDB can effectively reveal the relationships between circRNAs and HCC. As the first circRNAs database to provide analysis and comparison functions, it is of great significance for researchers to further study the rules of circRNAs, to understand the causes of circRNAs in disease discovery and to find target genes for therapeutic approaches.

A Novel Hybrid CNN-SVR for CRISPR/Cas9 Guide RNA Activity Prediction.

Accurate prediction of guide RNA (gRNA) on-target efficacy is critical for effective application of CRISPR/Cas9 system. Although some machine learning-based and convolutional neural network (CNN)-based methods have been proposed, prediction accuracy remains to be improved. Here, firstly we improved architectures of current CNNs for predicting gRNA on-target efficacy. Secondly, we proposed a novel hybrid system which combines our improved CNN with support vector regression (SVR). This CNN-SVR system is composed of two major components: a merged CNN as the front-end for extracting gRNA feature and an SVR as the back-end for regression and predicting gRNA cleavage efficiency. We demonstrate that CNN-SVR can effectively exploit features interactions from feed-forward directions to learn deeper features of gRNAs and their corresponding epigenetic features. Experiments on commonly used datasets show that our CNN-SVR system outperforms available state-of-the-art methods in terms of prediction accuracy, generalization, and robustness. Source codes are available at https://github.com/Peppags/CNN-SVR.

The relationship between DNA single-stranded damage response and double-stranded damage response.

The damage response of DNA single-stranded breaks(SSBs) and double-stranded breaks(DSBs) are two relatively independent processes involving different signaling pathways and protein factors, but there are still many overlapping parts. All of them can activate p53 protein, then the activated p53 regulates the damage response of single-stranded breaks or double-stranded breaks in transcriptional regulation and non-transcriptional regulation. Especially, the two types of damage would compete for RPA and ATR resources in damage repair process. The research has been focused on damage response of DNA single-stranded breaks or DNA double-stranded breaks. However, when single-stranded breaks and double-stranded breaks exist simultaneously, the DNA damage response remains to be elucidated. Here, we present a hybrid numerical model of p53 response and a hybrid numerical model of DNA damage repair exploring DNA damage repair and apoptosis mechanisms when DNA single-stranded breaks and DNA double-stranded breaks exist simultaneously. Firstly, when two kinds of damage are present at the same time, the response of p53 is graded, it means that p53 responds to single-stranded breaks preferentially; Secondly, DNA single-stranded breaks are repaired preferentially, and single-stranded breaks and double-stranded breaks can be repaired simultaneously after most of single-stranded breaks having been repaired; Moreover, single-stranded breaks are more likely to cause apoptosis, because the accumulation of p53 in DNA single-stranded breaks is faster than it in DNA double-stranded breaks and single-stranded breaks has lower threshold of apoptosis.

150-Gb/s SEFDM IM/DD transmission using log-MAP Viterbi decoding for short reach optical links.

Spectral efficient frequency division multiplexing (SEFDM) can improve the spectral efficiency for next-generation optical and wireless communications. In this work, we apply SEFDM in beyond 100-Gb/s optical intensity modulation and direct detection transmissions and propose a low-complexity logarithmic-maximum-a-posteriori (log-MAP) Viterbi decoding algorithm to achieve the maximum likelihood (ML) detection. We evaluate the likelihood of detections using a posteriori probability instead of Euclidean distance by taking both noise and inter-carrier interference into consideration. In order to balance the performance and complexity, we then employ Viterbi decoding principle to retain only certain paths with ML detections (a.k.a., the surviving paths) while discarding the others during the decoding procedure. Results show that the proposed log-MAP Viterbi decoding scheme achieves optimal performance due to the precise likelihood evaluation, which guarantees the retention of the global ML detection. By using the proposed decoding scheme, the data rate of SEFDM signals can reach 150-Gb/s in a 2-km standard single mode fiber transmission, using only 28-GHz bandwidth and 16-QAM modulation. Experimental results show that the 16-QAM modulated SEFDM signal with a bandwidth compression factor of 0.8 outperforms 32-QAM modulated OFDM, while both signals have the same bandwidth (28-GHz) and data rate (140-Gb/s), which demonstrate the superiority of SEFDM in optical short reach applications.

Dissecting Embryonic Stem Cell Self-Renewal and Differentiation Commitment from Quantitative Models.

To model quantitatively embryonic stem cell (ESC) self-renewal and differentiation by computational approaches, we developed a unified mathematical model for gene expression involved in cell fate choices. Our quantitative model comprised ESC master regulators and lineage-specific pivotal genes. It took the factors of multiple pathways as input and computed expression as a function of intrinsic transcription factors, extrinsic cues, epigenetic modifications, and antagonism between ESC master regulators and lineage-specific pivotal genes. In the model, the differential equations of expression of genes involved in cell fate choices from regulation relationship were established according to the transcription and degradation rates. We applied this model to the Murine ESC self-renewal and differentiation commitment and found that it modeled the expression patterns with good accuracy. Our model analysis revealed that Murine ESC was an attractor state in culture and differentiation was predominantly caused by antagonism between ESC master regulators and lineage-specific pivotal genes. Moreover, antagonism among lineages played a critical role in lineage reprogramming. Our results also uncovered that the ordered expression alteration of ESC master regulators over time had a central role in ESC differentiation fates. Our computational framework was generally applicable to most cell-type maintenance and lineage reprogramming.

DNA signals at isoform promoters.

Transcriptional heterogeneity is extensive in the genome, and most genes express variable transcript isoforms. However, whether variable transcript isoforms of one gene are regulated by common promoter elements remain to be elucidated. Here, we investigated whether isoform promoters of one gene have separated DNA signals for transcription and translation initiation. We found that TATA box and nucleosome-disfavored DNA sequences are prevalent in distinct transcript isoform promoters of one gene. These DNA signals are conserved among species. Transcript isoform has a RNA-determined unstructured region around its start site. We found that these DNA/RNA features facilitate isoform transcription and translation. These results suggest a DNA-encoded mechanism by which transcript isoform is generated.