Combining protein sequences and structures with transformers and equivariant graph neural networks to predict protein function.

MOTIVATION: Millions of protein sequences have been generated by numerous genome and transcriptome sequencing projects. However, experimentally determining the function of the proteins is still a time consuming, low-throughput, and expensive process, leading to a large protein sequence-function gap. Therefore, it is important to develop computational methods to accurately predict protein function to fill the gap. Even though many methods have been developed to use protein sequences as input to predict function, much fewer methods leverage protein structures in protein function prediction because there was lack of accurate protein structures for most proteins until recently. RESULTS: We developed TransFun-a method using a transformer-based protein language model and 3D-equivariant graph neural networks to distill information from both protein sequences and structures to predict protein function. It extracts feature embeddings from protein sequences using a pre-trained protein language model (ESM) via transfer learning and combines them with 3D structures of proteins predicted by AlphaFold2 through equivariant graph neural networks. Benchmarked on the CAFA3 test dataset and a new test dataset, TransFun outperforms several state-of-the-art methods, indicating that the language model and 3D-equivariant graph neural networks are effective methods to leverage protein sequences and structures to improve protein function prediction. Combining TransFun predictions and sequence similarity-based predictions can further increase prediction accuracy. AVAILABILITY AND IMPLEMENTATION: The source code of TransFun is available at https://github.com/jianlin-cheng/TransFun.

Physical violence during pregnancy in sub-Saharan Africa: why it matters and who are most susceptible?

OBJECTIVE: The study assessed the prevalence of physical violence against pregnant women and its associated factors in sub-Saharan Africa (SSA). DESIGN: We analysed cross-sectional data of 108971 women in sexual unions from the most recent Demographic and Health Surveys of 26 countries in SSA. The predictors of physical violence were examined using a multilevel binary logistic regression. All the results were presented as adjusted odds ratios (aORs) with their corresponding 95% confidence intervals (CIs). SETTING: Twenty-six countries in SSA. PARTICIPANTS: 108 971 women who had ever been pregnant. OUTCOME MEASURE: Physical violence during pregnancy. RESULTS: Physical violence was identified in 6.0% of pregnant women in SSA. The highest prevalence (14.0%) was reported in South Africa, while Burkina Faso recorded the lowest (2.1%). Women who had primary (aOR=1.26, 95% CI=1.15, 1.38) and secondary education (aOR=1.15, 95% CI=1.01,1.32); those who were cohabiting (aOR=1.21, 95% CI=1.11, 1.32); those who were working (aOR=1.17, 95% CI=1.08, 1.28); and those whose partners had primary (aOR=1.15, 95% CI=1.04, 1.28) and secondary education (aOR=1.14, 95% CI=1.01, 1.28) were more likely to experience physical violence during pregnancy compared with those who had no formal education; those who were married; those who were not working, and those whose partners had no formal education, respectively. Moreover, women whose partners consumed alcohol (aOR=2.37, 95% CI=2.20, 2.56); those who had parity of four or more (aOR=2.06, 95% CI=1.57, 2.72); and those who perceived intimate partner violence (IPV) as a culturally accepted norm (aOR=1.55, 95% CI=1.44, 1.67) had higher odds of experiencing physical violence during pregnancy compared to those whose partners did not consume alcohol, those with parity zero, and those who did not perceive IPV as culturally accepted, respectively. On the contrary, women who were aged 35-39, those who were of the richest wealth index, and those in rural areas had reduced odds of experiencing physical violence during pregnancy. CONCLUSION: Based on the findings, community leaders are encouraged to liaise with law enforcement agencies to strictly enforce laws on gender-based violence by prosecuting perpetrators of IPV against pregnant women as a deterrent. Also, intensifying education on what constitutes IPV and the potential consequences on the health of pregnant women, their children, and their families will be laudable. Improving the socioeconomic status of women may also help to eliminate IPV perpetration against women at their pregnancy stage.

Dissection of a Down syndrome-associated trisomy to separate the gene dosage-dependent and -independent effects of an extra chromosome.

As an aneuploidy, trisomy is associated with mammalian embryonic and postnatal abnormalities. Understanding the underlying mechanisms involved in mutant phenotypes is broadly important and may lead to new strategies to treat clinical manifestations in individuals with trisomies, such as trisomy 21 [Down syndrome (DS)]. Although increased gene dosage effects because of a trisomy may account for the mutant phenotypes, there is also the possibility that phenotypic consequences of a trisomy can arise because of the presence of a freely segregating extra chromosome with its own centromere, i.e. a 'free trisomy' independent of gene dosage effects. Presently, there are no reports of attempts to functionally separate these two types of effects in mammals. To fill this gap, here we describe a strategy that employed two new mouse models of DS, Ts65Dn;Df(17)2Yey/+ and Dp(16)1Yey/Df(16)8Yey. Both models carry triplications of the same 103 human chromosome 21 gene orthologs; however, only Ts65Dn;Df(17)2Yey/+ mice carry a free trisomy. Comparison of these models revealed the gene dosage-independent impacts of an extra chromosome at the phenotypic and molecular levels for the first time. They are reflected by impairments of Ts65Dn;Df(17)2Yey/+ males in T-maze tests when compared with Dp(16)1Yey/Df(16)8Yey males. Results from the transcriptomic analysis suggest the extra chromosome plays a major role in trisomy-associated expression alterations of disomic genes beyond gene dosage effects. This model system can now be used to deepen our mechanistic understanding of this common human aneuploidy and obtain new insights into the effects of free trisomies in other human diseases such as cancers.

Partner alcohol consumption and intimate partner violence among women in Papua New Guinea: a cross-sectional analysis of Demographic and Health Survey.

OBJECTIVE: We examined the association between partner alcohol consumption and the experience of intimate partner violence among women in Papua New Guinea. DESIGN: We performed a cross-sectional analyses of data extracted from the 2016-2018 Papua New Guinea Demographic and Health Survey. We included 3319 women in sexual unions. Multilevel binary logistic regression analysis was used to examine the association between partner alcohol consumption and intimate partner violence, controlling for the covariates. Results from the regression analysis were presented using the crude odds ratios (cORs) and adjusted odds ratios (aORs), with 95% confidence intervals (CIs). SETTING: Papua New Guinea. PARTICIPANTS: Women aged 15-49 years in sexual unions. OUTCOME MEASURES: Physical, emotional, and sexual violence. RESULTS: The prevalence of physical, emotional and sexual violence among women in sexual unions in Papua New Guinea were 45.9% (42.4 to 47.7), 45.1% (43.4 to 46.8) and 24.3% (22.9 to 25.8), respectively. The level of partner alcohol consumption was 57.3%. Women whose partners consumed alcohol were more likely to experience physical violence (aOR=2.86, 95% CI=2.43 to 3.37), emotional violence (aOR=2.89, 95% CI=2.44 to 3.43) and sexual violence (aOR=2.56, 95% CI=2.08 to 3.16) compared with those whose partners did not consume alcohol. CONCLUSION: This study found a relatively high prevalence of intimate partner violence among women in Papua New Guinea. Most importantly, this study found partner alcohol consumption to be significantly and positively associated with intimate partner violence. The study, therefore, recommends that interventions seeking to reduce intimate partner violence among women in Papua New Guinea should intensify behaviour change and education on reducing or eliminating partner alcohol consumption.

Combining protein sequences and structures with transformers and equivariant graph neural networks to predict protein function.

Motivation: Millions of protein sequences have been generated by numerous genome and transcriptome sequencing projects. However, experimentally determining the function of the proteins is still a time consuming, low-throughput, and expensive process, leading to a large protein sequence-function gap. Therefore, it is important to develop computational methods to accurately predict protein function to fill the gap. Even though many methods have been developed to use protein sequences as input to predict function, much fewer methods leverage protein structures in protein function prediction because there was lack of accurate protein structures for most proteins until recently. Results: We developed TransFun - a method using a transformer-based protein language model and 3D-equivariant graph neural networks to distill information from both protein sequences and structures to predict protein function. It extracts feature embeddings from protein sequences using a pre-trained protein language model (ESM) via transfer learning and combines them with 3D structures of proteins predicted by AlphaFold2 through equivariant graph neural networks. Benchmarked on the CAFA3 test dataset and a new test dataset, TransFun outperforms several state-of-the-art methods, indicating the language model and 3D-equivariant graph neural networks are effective methods to leverage protein sequences and structures to improve protein function prediction. Combining TransFun predictions and sequence similarity-based predictions can further increase prediction accuracy. Availability: The source code of TransFun is available at https://github.com/jianlin-cheng/TransFun. Contact: chengji@missouri.edu.

The counterpart congenital overgrowth syndromes Beckwith-Wiedemann Syndrome in human and large offspring syndrome in bovine involve alterations in DNA methylation, transcription, and chromatin configuration.

Beckwith-Wiedemann Syndrome (BWS, OMIM #130650) is a congenital epigenetic disorder in humans which affects approximately 1 in 10,340 children. The incidence is likely an underestimation as the condition is usually recognized based on observable phenotypes at birth. BWS children have up to a 28% risk of developing tumors and currently, only 80% of patients can be corroborated molecularly (epimutations/variants). It is unknown how the subtypes of this condition are molecularly similar/dissimilar globally, therefore there is a need to deeply characterize the syndrome at the molecular level. Here we characterize the methylome, transcriptome and chromatin configuration of 18 BWS individuals together with the animal model of the condition, the bovine large offspring syndrome (LOS). Sex specific comparisons are performed for a subset of the BWS patients and LOS. Given that this epigenetic overgrowth syndrome has been characterized as a loss-of-imprinting condition, parental allele-specific comparisons were performed using the bovine animal model. In general, the differentially methylated regions (DMRs) detected in BWS and LOS showed significant enrichment for CTCF binding sites. Altered chromosome compartments in BWS and LOS were positively correlated with gene expression changes, and the promoters of differentially expressed genes showed significant enrichment for DMRs, differential topologically associating domains, and differential A/B compartments in some comparisons of BWS subtypes and LOS. We show shared regions of dysregulation between BWS and LOS, including several HOX gene clusters, and also demonstrate that altered DNA methylation differs between the clinically epigenetically identified BWS patients and those identified as having DNA variants (i.e. CDKN1C microdeletion). Lastly, we highlight additional genes and genomic regions that have the potential to serve as targets for biomarker development to improve current molecular methodologies. In summary, our results suggest that genome-wide alternation of chromosome architecture, which is partially caused by DNA methylation changes, also contribute to the development of BWS and LOS.

Deep Learning Prediction of Severe Health Risks for Pediatric COVID-19 Patients with a Large Feature Set in 2021 BARDA Data Challenge.

Most children infected with COVID-19 have no or mild symptoms and can recover automatically by themselves, but some pediatric COVID-19 patients need to be hospitalized or even to receive intensive medical care (e.g., invasive mechanical ventilation or cardiovascular support) to recover from the illnesses. Therefore, it is critical to predict the severe health risk that COVID-19 infection poses to children to provide precise and timely medical care for vulnerable pediatric COVID-19 patients. However, predicting the severe health risk for COVID-19 patients including children remains a significant challenge because many underlying medical factors affecting the risk are still largely unknown. In this work, instead of searching for a small number of most useful features to make prediction, we design a novel large-scale bag-of-words like method to represent various medical conditions and measurements of COVID-19 patients. After some simple feature filtering based on logistical regression, the large set of features is used with a deep learning method to predict both the hospitalization risk for COVID-19 infected children and the severe complication risk for the hospitalized pediatric COVID-19 patients. The method was trained and tested the datasets of the Biomedical Advanced Research and Development Authority (BARDA) Pediatric COVID-19 Data Challenge held from Sept. 15 to Dec. 17, 2021. The results show that the approach can rather accurately predict the risk of hospitalization and severe complication for pediatric COVID-19 patients and deep learning is more accurate than other machine learning methods.

Allele-specific aberration of imprinted domain chromosome architecture associates with large offspring syndrome.

Large offspring syndrome (LOS) and Beckwith-Wiedemann syndrome are similar epigenetic congenital overgrowth conditions in ruminants and humans, respectively. We have reported global loss-of-imprinting, methylome epimutations, and gene misregulation in LOS. However, less than 4% of gene misregulation can be explained with short range (<20kb) alterations in DNA methylation. Therefore, we hypothesized that methylome epimutations in LOS affect chromosome architecture which results in misregulation of genes located at distances >20kb in cis and in trans (other chromosomes). Our analyses focused on two imprinted domains that frequently reveal misregulation in these syndromes, namely KvDMR1 and IGF2R. Using bovine fetal fibroblasts, we identified CTCF binding at IGF2R imprinting control region but not KvDMR1, and allele-specific chromosome architecture of these domains in controls. In LOS, analyses identified erroneous long-range contacts and clustering tendency in the direction of expression of misregulated genes. In conclusion, altered chromosome architecture is associated with LOS.

Risk factors associated with the coexistence of stunting, underweight, and wasting in children under 5 from 31 sub-Saharan African countries.

OBJECTIVE: This study investigated the risk factors associated with the coexistence of stunting, underweight, and wasting among children under age 5 in sub-Saharan Africa (SSA). DESIGN: Data of 127, 487 under-5 children from 31 countries in SSA were pooled from the Demographic and Health Surveys collected between 2010 and 2019. We examined the risk of coexistence of stunting, underweight, and wasting using multinomial logistic regression models. The results were presented using relative risk ratios (RRR) with corresponding confidence intervals (CIs). SETTING: Thirty-one sub-Saharan African countries. PARTICIPANTS: Children under age 5. OUTCOME MEASURES: The outcome variables were three child anthropometrics: stunting (height-for-age z-scores); underweight (weight-for-age z-scores) and wasting (weight-for-height z-scores). RESULTS: The prevalence of coexistence of stunting, underweight, and wasting varied across countries, with the highest (12.14%) and lowest (0.58%) prevalences of coexistence of stunting, underweight and wasting in Benin and Gambia respectively. The risk of coexistence of the three indicators of undernutrition was higher among children aged 1 year (RRR=3.714; 95% CI 3.319 to 4.156) compared with those aged 0. The risk of coexistence of the three dimensions was lower among female children (RRR=0.468 95% CI 0.420 to 0.51), but higher for those with small size at birth (RRR=3.818; CI 3.383 to 4.308), those whose mothers had no education (RRR=3.291; 95% CI 1.961 to 5.522), not working (RRR=1.195; 95% CI 1.086 to 1.314), had no antenatal visits during pregnancy (RRR=1.364; 95% CI 1.20 to 1.541), children delivered at home (RRR=1.372; CI 1.232 to 1.529), those from poor households (RRR=1.408; 95% CI 1.235 to 1.605), those whose mothers had no access to media (RRR=1.255; 95% CI 1.144 to 1.377) and living in households with an unimproved toilet facility (RRR=1.158; 95% CI 1.032 to 1.300). CONCLUSIONS: Findings suggest the urgent need for consideration of the coexistence of stunting, wasting and underweight among under-5 children in policy design and programming of interventions to eradicate child malnutrition in SSA. In the short-term, national-level policies and interventions need to be well tailored considering the compositional characteristics.