Uncovering Clinical Risk Factors and Predicting Severe COVID-19 Cases Using UK Biobank Data: Machine Learning Approach.

BACKGROUND: COVID-19 is a major public health concern. Given the extent of the pandemic, it is urgent to identify risk factors associated with disease severity. More accurate prediction of those at risk of developing severe infections is of high clinical importance. OBJECTIVE: Based on the UK Biobank (UKBB), we aimed to build machine learning models to predict the risk of developing severe or fatal infections, and uncover major risk factors involved. METHODS: We first restricted the analysis to infected individuals (n=7846), then performed analysis at a population level, considering those with no known infection as controls (ncontrols=465,728). Hospitalization was used as a proxy for severity. A total of 97 clinical variables (collected prior to the COVID-19 outbreak) covering demographic variables, comorbidities, blood measurements (eg, hematological/liver/renal function/metabolic parameters), anthropometric measures, and other risk factors (eg, smoking/drinking) were included as predictors. We also constructed a simplified (lite) prediction model using 27 covariates that can be more easily obtained (demographic and comorbidity data). XGboost (gradient-boosted trees) was used for prediction and predictive performance was assessed by cross-validation. Variable importance was quantified by Shapley values (ShapVal), permutation importance (PermImp), and accuracy gain. Shapley dependency and interaction plots were used to evaluate the pattern of relationships between risk factors and outcomes. RESULTS: A total of 2386 severe and 477 fatal cases were identified. For analyses within infected individuals (n=7846), our prediction model achieved area under the receiving-operating characteristic curve (AUC-ROC) of 0.723 (95% CI 0.711-0.736) and 0.814 (95% CI 0.791-0.838) for severe and fatal infections, respectively. The top 5 contributing factors (sorted by ShapVal) for severity were age, number of drugs taken (cnt_tx), cystatin C (reflecting renal function), waist-to-hip ratio (WHR), and Townsend deprivation index (TDI). For mortality, the top features were age, testosterone, cnt_tx, waist circumference (WC), and red cell distribution width. For analyses involving the whole UKBB population, AUCs for severity and fatality were 0.696 (95% CI 0.684-0.708) and 0.825 (95% CI 0.802-0.848), respectively. The same top 5 risk factors were identified for both outcomes, namely, age, cnt_tx, WC, WHR, and TDI. Apart from the above, age, cystatin C, TDI, and cnt_tx were among the top 10 across all 4 analyses. Other diseases top ranked by ShapVal or PermImp were type 2 diabetes mellitus (T2DM), coronary artery disease, atrial fibrillation, and dementia, among others. For the "lite" models, predictive performances were broadly similar, with estimated AUCs of 0.716, 0.818, 0.696, and 0.830, respectively. The top ranked variables were similar to above, including age, cnt_tx, WC, sex (male), and T2DM. CONCLUSIONS: We identified numerous baseline clinical risk factors for severe/fatal infection by XGboost. For example, age, central obesity, impaired renal function, multiple comorbidities, and cardiometabolic abnormalities may predispose to poorer outcomes. The prediction models may be useful at a population level to identify those susceptible to developing severe/fatal infections, facilitating targeted prevention strategies. A risk-prediction tool is also available online. Further replications in independent cohorts are required to verify our findings.

Genome-Wide Search for SNP Interactions in GWAS Data: Algorithm, Feasibility, Replication Using Schizophrenia Datasets.

In this study, we looked for potential gene-gene interaction in susceptibility to schizophrenia by an exhaustive searching for SNP-SNP interactions in 3 GWAS datasets (phs000021:phg000013, phs000021:phg000014, phs000167) using our recently published algorithm. The search space for SNP-SNP interaction was confined to 8 biologically plausible ways of interaction under dominant-dominant or recessive-recessive modes. First, we performed our search of all pair-wise combination of 729,454 SNPs after filtering by SNP genotype quality. All possible pairwise interactions of any 2 SNPs (5 × 1011) were exhausted to search for significant interaction which was defined by p-value of chi-square tests. Nine out the top 10 interactions, protein coding genes were partnered with non-coding RNA (ncRNA) which suggested a new alternative insight into interaction biology other than the frequently sought-after protein-protein interaction. Therefore, we extended to look for replication among the top 10,000 interaction SNP pairs and high proportion of concurrent genes forming the interaction pairs were found. The results indicated that an enrichment of signals over noise was present in the top 10,000 interactions. Then, replications of SNP-SNP interaction were confirmed for 14 SNPs-pairs in both replication datasets. Biological insight was highlighted by a potential binding between FHIT (protein coding gene) and LINC00969 (lncRNA) which showed a replicable interaction between their SNPs. Both of them were reported to have expression in brain. Our study represented an early attempt of exhaustive interaction analysis of GWAS data which also yield replicated interaction and new insight into understanding of genetic interaction in schizophrenia.

Epigenetic CRISPR Screens Identify Npm1 as a Therapeutic Vulnerability in Non-Small Cell Lung Cancer.

Despite advancements in treatment options, the overall cure and survival rates for non-small cell lung cancers (NSCLC) remain low. While small-molecule inhibitors of epigenetic regulators have recently emerged as promising cancer therapeutics, their application in patients with NSCLC is limited. To exploit epigenetic regulators as novel therapeutic targets in NSCLC, we performed pooled epigenome-wide CRISPR knockout screens in vitro and in vivo and identified the histone chaperone nucleophosmin 1 (Npm1) as a potential therapeutic target. Genetic ablation of Npm1 significantly attenuated tumor progression in vitro and in vivo. Furthermore, KRAS-mutant cancer cells were more addicted to NPM1 expression. Genetic ablation of Npm1 rewired the balance of metabolism in cancer cells from predominant aerobic glycolysis to oxidative phosphorylation and reduced the population of tumor-propagating cells. Overall, our results support NPM1 as a therapeutic vulnerability in NSCLC. SIGNIFICANCE: Epigenome-wide CRISPR knockout screens identify NPM1 as a novel metabolic vulnerability and demonstrate that targeting NPM1 is a new therapeutic opportunity for patients with NSCLC.

Exploring Diseases/Traits and Blood Proteins Causally Related to Expression of ACE2, the Putative Receptor of SARS-CoV-2: A Mendelian Randomization Analysis Highlights Tentative Relevance of Diabetes-Related Traits.

OBJECTIVE: COVID-19 has become a major public health problem. There is good evidence that ACE2 is a receptor for SARS-CoV-2, and high expression of ACE2 may increase susceptibility to infection. We aimed to explore risk factors affecting susceptibility to infection and prioritize drug repositioning candidates, based on Mendelian randomization (MR) studies on ACE2 lung expression. RESEARCH DESIGN AND METHODS: We conducted a phenome-wide MR study to prioritize diseases/traits and blood proteins causally linked to ACE2 lung expression in GTEx. We also explored drug candidates whose targets overlapped with the top-ranked proteins in MR, as these drugs may alter ACE2 expression and may be clinically relevant. RESULTS: The most consistent finding was tentative evidence of an association between diabetes-related traits and increased ACE2 expression. Based on one of the largest genome-wide association studies on type 2 diabetes mellitus (T2DM) to date (N = 898,130), T2DM was causally linked to raised ACE2 expression (P = 2.91E-03; MR-IVW). Significant associations (at nominal level; P < 0.05) with ACE2 expression were observed across multiple diabetes data sets and analytic methods for T1DM, T2DM, and related traits including early start of insulin. Other diseases/traits having nominal significant associations with increased expression included inflammatory bowel disease, (estrogen receptor-positive) breast cancer, lung cancer, asthma, smoking, and elevated alanine aminotransferase. We also identified drugs that may target the top-ranked proteins in MR, such as fostamatinib and zinc. CONCLUSIONS: Our analysis suggested that diabetes and related traits may increase ACE2 expression, which may influence susceptibility to infection (or more severe infection). However, none of these findings withstood rigorous multiple testing corrections (at false discovery rate <0.05). Proteome-wide MR analyses might help uncover mechanisms underlying ACE2 expression and guide drug repositioning. Further studies are required to verify our findings.

Implications of de novo mutations in guiding drug discovery: A study of four neuropsychiatric disorders.

Recent studies have suggested an important role of de novo mutations (DNMs) in neuropsychiatric disorders. As DNMs are not subject to elimination due to evolutionary pressure, they are likely to have greater disruptions on biological functions. While a number of sequencing studies have been performed on neuropsychiatric disorders, the implications of DNMs for drug discovery remain to be explored. In this study, we employed a gene-set analysis approach to address this issue. Four neuropsychiatric disorders were studied, including schizophrenia (SCZ), autistic spectrum disorders (ASD), intellectual disability (ID) and epilepsy. We first identified gene-sets associated with different drugs, and analyzed whether the gene-set pertaining to each drug overlaps with DNMs more than expected by chance. We also assessed which medication classes are enriched among the prioritized drugs. We discovered that neuropsychiatric drug classes were indeed significantly enriched for DNMs of all four disorders; in particular, antipsychotics and antiepileptics were the most strongly enriched drug classes for SCZ and epilepsy respectively. Interestingly, we revealed enrichment of several unexpected drug classes, such as lipid-lowering agents for SCZ and anti-neoplastic agents. By inspecting individual hits, we also uncovered other interesting drug candidates or mechanisms (e.g. histone deacetylase inhibition and retinoid signaling) that might warrant further investigations. Taken together, this study provided evidence for the usefulness of DNMs in guiding drug discovery or repositioning.

Risk prediction of complex diseases from family history and known susceptibility loci, with applications for cancer screening.

Risk prediction based on genomic profiles has raised a lot of attention recently. However, family history is usually ignored in genetic risk prediction. In this study we proposed a statistical framework for risk prediction given an individual's genotype profile and family history. Genotype information about the relatives can also be incorporated. We allow risk prediction given the current age and follow-up period and consider competing risks of mortality. The framework allows easy extension to any family size and structure. In addition, the predicted risk at any percentile and the risk distribution graphs can be computed analytically. We applied the method to risk prediction for breast and prostate cancers by using known susceptibility loci from genome-wide association studies. For breast cancer, in the population the 10-year risk at age 50 ranged from 1.1% at the 5th percentile to 4.7% at the 95th percentile. If we consider the average 10-year risk at age 50 (2.39%) as the threshold for screening, the screening age ranged from 62 at the 20th percentile to 38 at the 95th percentile (and some never reach the threshold). For women with one affected first-degree relative, the 10-year risks ranged from 2.6% (at the 5th percentile) to 8.1% (at the 95th percentile). For prostate cancer, the corresponding 10-year risks at age 60 varied from 1.8% to 14.9% in the population and from 4.2% to 23.2% in those with an affected first-degree relative. We suggest that for some diseases genetic testing that incorporates family history can stratify people into diverse risk categories and might be useful in targeted prevention and screening.

Uncovering the total heritability explained by all true susceptibility variants in a genome-wide association study.

Genome-wide association studies (GWAS) have become increasingly popular recently and contributed to the discovery of many susceptibility variants. However, a large proportion of the heritability still remained unexplained. This observation raises queries regarding the ability of GWAS to uncover the genetic basis of complex diseases. In this study, we propose a simple and fast statistical framework to estimate the total heritability explained by all true susceptibility variants in a GWAS. It is expected that many true risk variants will not be detected in a GWAS due to limited power. The proposed framework aims at recovering the "hidden" heritability. Importantly, only the summary z-statistics are required as input and no raw genotype data are needed. The strategy is to recover the true effect sizes from the observed z-statistics. The methodology does not rely on any distributional assumptions of the effect sizes of variants. Both binary and quantitative traits can be handled and covariates may be included. Population-based or family-based designs are allowed as long as the summary statistics are available. Simulations were conducted and showed satisfactory performance of the proposed approach. Application to real data (Crohn's disease, HDL, LDL, and triglycerides) reveals that at least around 10-20% of variance in liability or phenotype can be explained by GWAS panels. This translates to around 10-40% of the total heritability for the studied traits.

Evaluating the heritability explained by known susceptibility variants: a survey of ten complex diseases.

Recently, an increasing number of susceptibility variants have been identified for complex diseases. At the same time, the concern of "missing heritability" has also emerged. There is however no unified way to assess the heritability explained by individual genetic variants for binary outcomes. A systemic and quantitative assessment of the degree of "missing heritability" for complex diseases is lacking. In this study, we measure the variance in liability explained by individual variants, which can be directly interpreted as the locus-specific heritability. The method is extended to deal with haplotypes, multi-allelic markers, multi-locus genotypes, and markers in linkage disequilibrium. Methods to estimate the standard error and confidence interval are proposed. To assess our current level of understanding of the genetic basis of complex diseases, we conducted a survey of 10 diseases, evaluating the total variance explained by the known variants. The diseases under evaluation included Alzheimer's disease, bipolar disorder, breast cancer, coronary artery disease, Crohn's disease, prostate cancer, schizophrenia, systemic lupus erythematosus (SLE), type 1 diabetes and type 2 diabetes. The median total variance explained across the 10 diseases was 9.81%, while the median variance explained per associated SNP was around 0.25%. Our results suggest that a substantial proportion of heritability remains unexplained for the diseases under study. Programs to implement the methodologies described in this paper are available at http://sites.google.com/site/honcheongso/software/varexp.

Estimating the total number of susceptibility variants underlying complex diseases from genome-wide association studies.

Recently genome-wide association studies (GWAS) have identified numerous susceptibility variants for complex diseases. In this study we proposed several approaches to estimate the total number of variants underlying these diseases. We assume that the variance explained by genetic markers (Vg) follow an exponential distribution, which is justified by previous studies on theories of adaptation. Our aim is to fit the observed distribution of Vg from GWAS to its theoretical distribution. The number of variants is obtained by the heritability divided by the estimated mean of the exponential distribution. In practice, due to limited sample sizes, there is insufficient power to detect variants with small effects. Therefore the power was taken into account in fitting. Besides considering the most significant variants, we also tried to relax the significance threshold, allowing more markers to be fitted. The effects of false positive variants were removed by considering the local false discovery rates. In addition, we developed an alternative approach by directly fitting the z-statistics from GWAS to its theoretical distribution. In all cases, the "winner's curse" effect was corrected analytically. Confidence intervals were also derived. Simulations were performed to compare and verify the performance of different estimators (which incorporates various means of winner's curse correction) and the coverage of the proposed analytic confidence intervals. Our methodology only requires summary statistics and is able to handle both binary and continuous traits. Finally we applied the methods to a few real disease examples (lipid traits, type 2 diabetes and Crohn's disease) and estimated that hundreds to nearly a thousand variants underlie these traits.

Effect size measures in genetic association studies and age-conditional risk prediction.

The interest in risk prediction using genomic profiles has surged recently. A proper interpretation of effect size measures in association studies is crucial to accurate risk prediction. In this study, we clarified the relationship between the odds ratio (OR), relative risk and incidence rate ratios in the context of genetic association studies. We demonstrated that under the common practice of sampling prevalent cases and controls, the resulting ORs approximate the incidence rate ratios. Based on this result, we presented a framework to compute the disease risk given the current age and follow-up period (including lifetime risk), with consideration of competing risks of mortality. We considered two extensions. One is correcting the incidence rate to reflect the person-years alive and disease-free, the other is converting prevalence to incidence estimates. The methodology was applied to an example of breast cancer prediction. We observed that simply multiplying the OR by the average lifetime risk estimates yielded a final estimate >100% (101%), while using our method that accounts for competing risks produces an estimate of 63% only. We also applied the method to risk prediction of Alzheimer's disease in Hong Kong. We recommend that companies offering direct-to-consumer genetic testing employ more rigorous prediction algorithms considering competing risks.