Data-driven identification of predictive risk biomarkers for subgroups of osteoarthritis using interpretable machine learning.

Osteoarthritis (OA) is increasing in prevalence and has a severe impact on patients' lives. However, our understanding of biomarkers driving OA risk remains limited. We developed a model predicting the five-year risk of OA diagnosis, integrating retrospective clinical, lifestyle and biomarker data from the UK Biobank (19,120 patients with OA, ROC-AUC: 0.72, 95%CI (0.71-0.73)). Higher age, BMI and prescription of non-steroidal anti-inflammatory drugs contributed most to increased OA risk prediction ahead of diagnosis. We identified 14 subgroups of OA risk profiles. These subgroups were validated in an independent set of patients evaluating the 11-year OA risk, with 88% of patients being uniquely assigned to one of the 14 subgroups. Individual OA risk profiles were characterised by personalised biomarkers. Omics integration demonstrated the predictive importance of key OA genes and pathways (e.g., GDF5 and TGF-ß signalling) and OA-specific biomarkers (e.g., CRTAC1 and COL9A1). In summary, this work identifies opportunities for personalised OA prevention and insights into its underlying pathogenesis.

Characteristics of white blood cell count in acute lymphoblastic leukemia: A COST LEGEND phenotype-genotype study.

BACKGROUND: White blood cell count (WBC) as a measure of extramedullary leukemic cell survival is a well-known prognostic factor in acute lymphoblastic leukemia (ALL), but its biology, including impact of host genome variants, is poorly understood. METHODS: We included patients treated with the Nordic Society of Paediatric Haematology and Oncology (NOPHO) ALL-2008 protocol (N = 2347, 72% were genotyped by Illumina Omni2.5exome-8-Bead chip) aged 1-45 years, diagnosed with B-cell precursor (BCP-) or T-cell ALL (T-ALL) to investigate the variation in WBC. Spline functions of WBC were fitted correcting for association with age across ALL subgroups of immunophenotypes and karyotypes. The residuals between spline WBC and actual WBC were used to identify WBC-associated germline genetic variants in a genome-wide association study (GWAS) while adjusting for age and ALL subtype associations. RESULTS: We observed an overall inverse correlation between age and WBC, which was stronger for the selected patient subgroups of immunophenotype and karyotypes (ρBCP-ALL  = -.17, ρT-ALL  = -.19; p < 3 × 10-4 ). Spline functions fitted to age, immunophenotype, and karyotype explained WBC variation better than age alone (ρ = .43, p << 2 × 10-6 ). However, when the spline-adjusted WBC residuals were used as phenotype, no GWAS significant associations were found. Based on available annotation, the top 50 genetic variants suggested effects on signal transduction, translation initiation, cell development, and proliferation. CONCLUSION: These results indicate that host genome variants do not strongly influence WBC across ALL subsets, and future studies of why some patients are more prone to hyperleukocytosis should be performed within specific ALL subsets that apply more complex analyses to capture potential germline variant interactions and impact on WBC.

Acute central nervous system toxicity during treatment of pediatric acute lymphoblastic leukemia: phenotypes, risk factors and genotypes.

Central nervous system (CNS) toxicity is common at diagnosis and during treatment of pediatric acute lymphoblastic leukemia (ALL). We studied CNS toxicity in 1,464 children aged 1.0-17.9 years, diagnosed with ALL and treated according to the Nordic Society of Pediatric Hematology and Oncology ALL2008 protocol. Genome-wide association studies, and a candidate single-nucleotide polymorphism (SNP; n=19) study were performed in 1,166 patients. Findings were validated in an independent Australian cohort of children with ALL (n=797) in whom two phenotypes were evaluated: diverse CNS toxicities (n=103) and methotrexate-related CNS toxicity (n=48). In total, 135/1,464 (9.2%) patients experienced CNS toxicity for a cumulative incidence of 8.7% (95% confidence interval: 7.31-10.20) at 12 months from diagnosis. Patients aged ≥10 years had a higher risk of CNS toxicity than had younger patients (16.3% vs. 7.4%; P<0.001). The most common CNS toxicities were posterior reversible encephalopathy syndrome (n=52, 43 with seizures), sinus venous thrombosis (n=28, 9 with seizures), and isolated seizures (n=16). The most significant SNP identified by the genome-wide association studies did not reach genomic significance (lowest P-value: 1.11x10-6), but several were annotated in genes regulating neuronal functions. In candidate SNP analysis, ATXN1 rs68082256, related to epilepsy, was associated with seizures in patients <10 years (P=0.01). ATXN1 rs68082256 was validated in the Australian cohort with diverse CNS toxicities (P=0.04). The role of ATXN1 as well as the novel SNP in neurotoxicity in pediatric ALL should be further explored.

Can Machine Learning Models Predict Asparaginase-associated Pancreatitis in Childhood Acute Lymphoblastic Leukemia.

Asparaginase-associated pancreatitis (AAP) frequently affects children treated for acute lymphoblastic leukemia (ALL) causing severe acute and persisting complications. Known risk factors such as asparaginase dosing, older age and single nucleotide polymorphisms (SNPs) have insufficient odds ratios to allow personalized asparaginase therapy. In this study, we explored machine learning strategies for prediction of individual AAP risk. We integrated information on age, sex, and SNPs based on Illumina Omni2.5exome-8 arrays of patients with childhood ALL (N=1564, 244 with AAP 1.0 to 17.9 yo) from 10 international ALL consortia into machine learning models including regression, random forest, AdaBoost and artificial neural networks. A model with only age and sex had area under the receiver operating characteristic curve (ROC-AUC) of 0.62. Inclusion of 6 pancreatitis candidate gene SNPs or 4 validated pancreatitis SNPs boosted ROC-AUC somewhat (0.67) while 30 SNPs, identified through our AAP genome-wide association study cohort, boosted performance (0.80). Most predictive features included rs10273639 (PRSS1-PRSS2), rs10436957 (CTRC), rs13228878 (PRSS1/PRSS2), rs1505495 (GALNTL6), rs4655107 (EPHB2) and age (1 to 7 y). Second AAP following asparaginase re-exposure was predicted with ROC-AUC: 0.65. The machine learning models assist individual-level risk assessment of AAP for future prevention trials, and may legitimize asparaginase re-exposure when AAP risk is predicted to be low.

Large-Scale Functional Genomics Screen to Identify Modulators of Human ß-Cell Insulin Secretion.

Type 2 diabetes (T2D) is a chronic metabolic disorder affecting almost half a billion people worldwide. Impaired function of pancreatic ß-cells is both a hallmark of T2D and an underlying factor in the pathophysiology of the disease. Understanding the cellular mechanisms regulating appropriate insulin secretion has been of long-standing interest in the scientific and clinical communities. To identify novel genes regulating insulin secretion we developed a robust arrayed siRNA screen measuring basal, glucose-stimulated, and augmented insulin secretion by EndoC-ßH1 cells, a human ß-cell line, in a 384-well plate format. We screened 521 candidate genes selected by text mining for relevance to T2D biology and identified 23 positive and 68 negative regulators of insulin secretion. Among these, we validated ghrelin receptor (GHSR), and two genes implicated in endoplasmic reticulum stress, ATF4 and HSPA5. Thus, we have demonstrated the feasibility of using EndoC-ßH1 cells for large-scale siRNA screening to identify candidate genes regulating ß-cell insulin secretion as potential novel drug targets. Furthermore, this screening format can be adapted to other disease-relevant functional endpoints to enable large-scale screening for targets regulating cellular mechanisms contributing to the progressive loss of functional ß-cell mass occurring in T2D.

Data integration for prediction of weight loss in randomized controlled dietary trials.

Diet is an important component in weight management strategies, but heterogeneous responses to the same diet make it difficult to foresee individual weight-loss outcomes. Omics-based technologies now allow for analysis of multiple factors for weight loss prediction at the individual level. Here, we classify weight loss responders (N = 106) and non-responders (N = 97) of overweight non-diabetic middle-aged Danes to two earlier reported dietary trials over 8 weeks. Random forest models integrated gut microbiome, host genetics, urine metabolome, measures of physiology and anthropometrics measured prior to any dietary intervention to identify individual predisposing features of weight loss in combination with diet. The most predictive models for weight loss included features of diet, gut bacterial species and urine metabolites (ROC-AUC: 0.84-0.88) compared to a diet-only model (ROC-AUC: 0.62). A model ensemble integrating multi-omics identified 64% of the non-responders with 80% confidence. Such models will be useful to assist in selecting appropriate weight management strategies, as individual predisposition to diet response varies.