Genome interpretation in a federated learning context allows the multi-center exome-based risk prediction of Crohn's disease patients.

High-throughput sequencing allowed the discovery of many disease variants, but nowadays it is becoming clear that the abundance of genomics data mostly just moved the bottleneck in Genetics and Precision Medicine from a data availability issue to a data interpretation issue. To solve this empasse it would be beneficial to apply the latest Deep Learning (DL) methods to the Genome Interpretation (GI) problem, similarly to what AlphaFold did for Structural Biology. Unfortunately DL requires large datasets to be viable, and aggregating genomics datasets poses several legal, ethical and infrastructural complications. Federated Learning (FL) is a Machine Learning (ML) paradigm designed to tackle these issues. It allows ML methods to be collaboratively trained and tested on collections of physically separate datasets, without requiring the actual centralization of sensitive data. FL could thus be key to enable DL applications to GI on sufficiently large genomics data. We propose FedCrohn, a FL GI Neural Network model for the exome-based Crohn's Disease risk prediction, providing a proof-of-concept that FL is a viable paradigm to build novel ML GI approaches. We benchmark it in several realistic scenarios, showing that FL can indeed provide performances similar to conventional ML on centralized data, and that collaborating in FL initiatives is likely beneficial for most of the medical centers participating in them.

Augmenting THerapeutic Effectiveness Through Novel Analytics (ATHENA) - A Public and Private Partnership Project Funded by the Flemish Government (VLAIO).

The complexity and heterogeneity of cancers leads to variable responses of patients to treatments and interventions. Developing models that accurately predict patient's care pathways using prognostic and predictive biomarkers is increasingly important in both clinical practice and scientific research. The main objective of the ATHENA project is to: (1) accelerate data driven precision medicine for two use cases - bladder cancer and multiple myeloma, (2) apply distributed and privacy-preserving analytical methods/ algorithms to stratify patients (decision support), (3) help healthcare professionals deliver earlier and better targeted treatments, and (4) explore care pathway automations and improve outcomes for each patient. Challenges associated with data sharing and integration will be addressed and an appropriate federated data ecosystem will be created, enabling an interoperable foundation for data exchange, analysis and interpretation. By combining multidisciplinary expertise and tackling knowledge gaps in ATHENA, we propose a novel federated privacy preserving platform for oncology research.

Current cancer driver variant predictors learn to recognize driver genes instead of functional variants.

BACKGROUND: Identifying variants that drive tumor progression (driver variants) and distinguishing these from variants that are a byproduct of the uncontrolled cell growth in cancer (passenger variants) is a crucial step for understanding tumorigenesis and precision oncology. Various bioinformatics methods have attempted to solve this complex task. RESULTS: In this study, we investigate the assumptions on which these methods are based, showing that the different definitions of driver and passenger variants influence the difficulty of the prediction task. More importantly, we prove that the data sets have a construction bias which prevents the machine learning (ML) methods to actually learn variant-level functional effects, despite their excellent performance. This effect results from the fact that in these data sets, the driver variants map to a few driver genes, while the passenger variants spread across thousands of genes, and thus just learning to recognize driver genes provides almost perfect predictions. CONCLUSIONS: To mitigate this issue, we propose a novel data set that minimizes this bias by ensuring that all genes covered by the data contain both driver and passenger variants. As a result, we show that the tested predictors experience a significant drop in performance, which should not be considered as poorer modeling, but rather as correcting unwarranted optimism. Finally, we propose a weighting procedure to completely eliminate the gene effects on such predictions, thus precisely evaluating the ability of predictors to model the functional effects of single variants, and we show that indeed this task is still open.

Insight into the protein solubility driving forces with neural attention.

Protein solubility is a key aspect for many biotechnological, biomedical and industrial processes, such as the production of active proteins and antibodies. In addition, understanding the molecular determinants of the solubility of proteins may be crucial to shed light on the molecular mechanisms of diseases caused by aggregation processes such as amyloidosis. Here we present SKADE, a novel Neural Network protein solubility predictor and we show how it can provide novel insight into the protein solubility mechanisms, thanks to its neural attention architecture. First, we show that SKADE positively compares with state of the art tools while using just the protein sequence as input. Then, thanks to the neural attention mechanism, we use SKADE to investigate the patterns learned during training and we analyse its decision process. We use this peculiarity to show that, while the attention profiles do not correlate with obvious sequence aspects such as biophysical properties of the aminoacids, they suggest that N- and C-termini are the most relevant regions for solubility prediction and are predictive for complex emergent properties such as aggregation-prone regions involved in beta-amyloidosis and contact density. Moreover, SKADE is able to identify mutations that increase or decrease the overall solubility of the protein, allowing it to be used to perform large scale in-silico mutagenesis of proteins in order to maximize their solubility.

Structure-based drug repositioning explains ibrutinib as VEGFR2 inhibitor.

Many drugs are promiscuous and bind to multiple targets. On the one hand, these targets may be linked to unwanted side effects, but on the other, they may achieve a combined desired effect (polypharmacology) or represent multiple diseases (drug repositioning). With the growth of 3D structures of drug-target complexes, it is today possible to study drug promiscuity at the structural level and to screen vast amounts of drug-target interactions to predict side effects, polypharmacological potential, and repositioning opportunities. Here, we pursue such an approach to identify drugs inactivating B-cells, whose dysregulation can function as a driver of autoimmune diseases. Screening over 500 kinases, we identified 22 candidate targets, whose knock out impeded the activation of B-cells. Among these 22 is the gene KDR, whose gene product VEGFR2 is a prominent cancer target with anti-VEGFR2 drugs on the market for over a decade. The main result of this paper is that structure-based drug repositioning for the identified kinase targets identified the cancer drug ibrutinib as micromolar VEGFR2 inhibitor with a very high therapeutic index in B-cell inactivation. These findings prove that ibrutinib is not only acting on the Bruton's tyrosine kinase BTK, against which it was designed. Instead, it may be a polypharmacological drug, which additionally targets angiogenesis via inhibition of VEGFR2. Therefore ibrutinib carries potential to treat other VEGFR2 associated disease. Structure-based drug repositioning explains ibrutinib's anti VEGFR2 action through the conservation of a specific pattern of interactions of the drug with BTK and VEGFR2. Overall, structure-based drug repositioning was able to predict these findings at a fraction of the time and cost of a conventional screen.

Exploring the limitations of biophysical propensity scales coupled with machine learning for protein sequence analysis.

Machine learning (ML) is ubiquitous in bioinformatics, due to its versatility. One of the most crucial aspects to consider while training a ML model is to carefully select the optimal feature encoding for the problem at hand. Biophysical propensity scales are widely adopted in structural bioinformatics because they describe amino acids properties that are intuitively relevant for many structural and functional aspects of proteins, and are thus commonly used as input features for ML methods. In this paper we reproduce three classical structural bioinformatics prediction tasks to investigate the main assumptions about the use of propensity scales as input features for ML methods. We investigate their usefulness with different randomization experiments and we show that their effectiveness varies among the ML methods used and the tasks. We show that while linear methods are more dependent on the feature encoding, the specific biophysical meaning of the features is less relevant for non-linear methods. Moreover, we show that even among linear ML methods, the simpler one-hot encoding can surprisingly outperform the "biologically meaningful" scales. We also show that feature selection performed with non-linear ML methods may not be able to distinguish between randomized and "real" propensity scales by properly prioritizing to the latter. Finally, we show that learning problem-specific embeddings could be a simple, assumptions-free and optimal way to perform feature learning/engineering for structural bioinformatics tasks.

Author Correction: Pan-cancer analysis of homozygous deletions in primary tumours uncovers rare tumour suppressors.

The original version of this Article omitted a declaration from the competing interests statement, which should have included the following: 'K.P.W. is President of Tempus Lab, Inc., Chicago, IL, USA'. This has now been corrected in both the PDF and HTML versions of the Article.

Author Correction: Pan-cancer analysis of homozygous deletions in primary tumours uncovers rare tumour suppressors.

The original version of this Article contained an error in the author affiliations. The affiliation of Kevin P. White with Tempus Labs, Inc., Chicago, IL, USA was inadvertently omitted.This has now been corrected in both the PDF and HTML versions of the Article.

Linking drug target and pathway activation for effective therapy using multi-task learning.

Despite the abundance of large-scale molecular and drug-response data, the insights gained about the mechanisms underlying treatment efficacy in cancer has been in general limited. Machine learning algorithms applied to those datasets most often are used to provide predictions without interpretation, or reveal single drug-gene association and fail to derive robust insights. We propose to use Macau, a bayesian multitask multi-relational algorithm to generalize from individual drugs and genes and explore the interactions between the drug targets and signaling pathways' activation. A typical insight would be: "Activation of pathway Y will confer sensitivity to any drug targeting protein X". We applied our methodology to the Genomics of Drug Sensitivity in Cancer (GDSC) screening, using gene expression of 990 cancer cell lines, activity scores of 11 signaling pathways derived from the tool PROGENy as cell line input and 228 nominal targets for 265 drugs as drug input. These interactions can guide a tissue-specific combination treatment strategy, for example suggesting to modulate a certain pathway to maximize the drug response for a given tissue. We confirmed in literature drug combination strategies derived from our result for brain, skin and stomach tissues. Such an analysis of interactions across tissues might help target discovery, drug repurposing and patient stratification strategies.

Repurposing High-Throughput Image Assays Enables Biological Activity Prediction for Drug Discovery.

In both academia and the pharmaceutical industry, large-scale assays for drug discovery are expensive and often impractical, particularly for the increasingly important physiologically relevant model systems that require primary cells, organoids, whole organisms, or expensive or rare reagents. We hypothesized that data from a single high-throughput imaging assay can be repurposed to predict the biological activity of compounds in other assays, even those targeting alternate pathways or biological processes. Indeed, quantitative information extracted from a three-channel microscopy-based screen for glucocorticoid receptor translocation was able to predict assay-specific biological activity in two ongoing drug discovery projects. In these projects, repurposing increased hit rates by 50- to 250-fold over that of the initial project assays while increasing the chemical structure diversity of the hits. Our results suggest that data from high-content screens are a rich source of information that can be used to predict and replace customized biological assays.

Pan-cancer analysis of homozygous deletions in primary tumours uncovers rare tumour suppressors.

Homozygous deletions are rare in cancers and often target tumour suppressor genes. Here, we build a compendium of 2218 primary tumours across 12 human cancer types and systematically screen for homozygous deletions, aiming to identify rare tumour suppressors. Our analysis defines 96 genomic regions recurrently targeted by homozygous deletions. These recurrent homozygous deletions occur either over tumour suppressors or over fragile sites, regions of increased genomic instability. We construct a statistical model that separates fragile sites from regions showing signatures of positive selection for homozygous deletions and identify candidate tumour suppressors within those regions. We find 16 established tumour suppressors and propose 27 candidate tumour suppressors. Several of these genes (including MGMT, RAD17, and USP44) show prior evidence of a tumour suppressive function. Other candidate tumour suppressors, such as MAFTRR, KIAA1551, and IGF2BP2, are novel. Our study demonstrates how rare tumour suppressors can be identified through copy number meta-analysis.

Genome-Wide Copy Number Variation Scan Identifies Complement Component C4 as Novel Susceptibility Gene for Crohn's Disease.

BACKGROUND: The genetic component of Crohn's disease (CD) is well known, with 140 susceptibility loci identified so far. In addition to single nucleotide polymorphisms typically studied in genome-wide scans, copy number variation is responsible for a large proportion of human genetic variation. METHODS: We performed a genome-wide search for copy number variants associated with CD using array comparative genomic hybridization. One of the found regions was validated independently through real-time PCR. Serum levels of the found gene were measured in patients and control subjects. RESULTS: We found copy number differences for the C4S and C4L gene variants of complement component C4 in the central major histocompatibility complex region on chromosome 6p21. Specifically, we saw that CD patients tend to have lower C4L and higher C4S copies than control subjects (P = 5.00 × 10 and P = 9.11 × 10), which was independent of known associated classical HLA I and II alleles (P = 7.68 × 10 and P = 6.29 × 10). Although C4 serum levels were not different between patients and control subjects, the relationship between C4 copy number and serum level was different for patients and control subjects with higher copy numbers leading to higher serum concentrations in control subjects, compared with CD patients (P < 0.001). CONCLUSIONS: C4 is part of the classical activation pathway of the complement system, which is important for (auto)immunity. Low C4L or high C4S copy number, and corresponding effects on C4 serum level, could lead to an exaggerated response against infections, possibly leading to (auto)immune disease.

In vitro screening of embryos by whole-genome sequencing: now, in the future or never?

STUDY QUESTION: What are the analytical and clinical validity and the clinical utility of in vitro screening of embryos by whole-genome sequencing? SUMMARY ANSWER: At present there are still many limitations in terms of analytical and clinical validity and utility and many ethical questions remain. WHAT IS KNOWN ALREADY: Whole-genome sequencing of IVF/ICSI embryos is technically possible. Many loss-of-function mutations exist in the general population without serious effects on the phenotype of the individual. Moreover, annotations of genes and the reference genome are still not 100% correct. STUDY DESIGN, SIZE, DURATION: We used publicly available samples from the 1000 Genomes project and Complete Genomics, together with 42 samples from in-house research samples of parents from trios to investigate the presence of loss-of-function mutations in healthy individuals. PARTICIPANTS/MATERIALS, SETTING, METHODS: In the samples, we looked for mutations in genes that are associated with a selection of severe Mendelian disorders with a known molecular basis. We looked for mutations predicted to be damaging by PolyPhen and SIFT and for mutations annotated as disease causing in Human Genome Mutation Database (HGMD). MAIN RESULTS AND THE ROLE OF CHANCE: More than 40% of individuals who can be considered healthy have mutations that are predicted to be damaging in genes associated with severe Mendelian disorders or are annotated as disease causing. LIMITATIONS, REASONS FOR CAUTION: The analysis relies on current knowledge and databases are continuously updated to reflect our increasing knowledge about the genome. In the process of our analysis several updates were already made. WIDER IMPLICATIONS OF THE FINDINGS: At this moment it is not advisable to use whole-genome sequencing as a tool to set up health profiles to select embryos for transfer. We also raise some ethical questions that have to be addressed before this technology can be used for embryo selection. TRIAL REGISTRATION NUMBER: N/A.

The genetic heterogeneity and mutational burden of engineered melanomas in zebrafish models.

BACKGROUND: Melanoma is the most deadly form of skin cancer. Expression of oncogenic BRAF or NRAS, which are frequently mutated in human melanomas, promote the formation of nevi but are not sufficient for tumorigenesis. Even with germline mutated p53, these engineered melanomas present with variable onset and pathology, implicating additional somatic mutations in a multi-hit tumorigenic process. RESULTS: To decipher the genetics of these melanomas, we sequence the protein coding exons of 53 primary melanomas generated from several BRAF(V600E) or NRAS(Q61K) driven transgenic zebrafish lines. We find that engineered zebrafish melanomas show an overall low mutation burden, which has a strong, inverse association with the number of initiating germline drivers. Although tumors reveal distinct mutation spectrums, they show mostly C > T transitions without UV light exposure, and enrichment of mutations in melanogenesis, p53 and MAPK signaling. Importantly, a recurrent amplification occurring with pre-configured drivers BRAF(V600E) and p53-/- suggests a novel path of BRAF cooperativity through the protein kinase A pathway. CONCLUSION: This is the first analysis of a melanoma mutational landscape in the absence of UV light, where tumors manifest with remarkably low mutation burden and high heterogeneity. Genotype specific amplification of protein kinase A in cooperation with BRAF and p53 mutation suggests the involvement of melanogenesis in these tumors. This work is important for defining the spectrum of events in BRAF or NRAS driven melanoma in the absence of UV light, and for informed exploitation of models such as transgenic zebrafish to better understand mechanisms leading to human melanoma formation.

Single-cell paired-end genome sequencing reveals structural variation per cell cycle.

The nature and pace of genome mutation is largely unknown. Because standard methods sequence DNA from populations of cells, the genetic composition of individual cells is lost, de novo mutations in cells are concealed within the bulk signal and per cell cycle mutation rates and mechanisms remain elusive. Although single-cell genome analyses could resolve these problems, such analyses are error-prone because of whole-genome amplification (WGA) artefacts and are limited in the types of DNA mutation that can be discerned. We developed methods for paired-end sequence analysis of single-cell WGA products that enable (i) detecting multiple classes of DNA mutation, (ii) distinguishing DNA copy number changes from allelic WGA-amplification artefacts by the discovery of matching aberrantly mapping read pairs among the surfeit of paired-end WGA and mapping artefacts and (iii) delineating the break points and architecture of structural variants. By applying the methods, we capture DNA copy number changes acquired over one cell cycle in breast cancer cells and in blastomeres derived from a human zygote after in vitro fertilization. Furthermore, we were able to discover and fine-map a heritable inter-chromosomal rearrangement t(1;16)(p36;p12) by sequencing a single blastomere. The methods will expedite applications in basic genome research and provide a stepping stone to novel approaches for clinical genetic diagnosis.

Meander: visually exploring the structural variome using space-filling curves.

The introduction of next generation sequencing methods in genome studies has made it possible to shift research from a gene-centric approach to a genome wide view. Although methods and tools to detect single nucleotide polymorphisms are becoming more mature, methods to identify and visualize structural variation (SV) are still in their infancy. Most genome browsers can only compare a given sequence to a reference genome; therefore, direct comparison of multiple individuals still remains a challenge. Therefore, the implementation of efficient approaches to explore and visualize SVs and directly compare two or more individuals is desirable. In this article, we present a visualization approach that uses space-filling Hilbert curves to explore SVs based on both read-depth and pair-end information. An interactive open-source Java application, called Meander, implements the proposed methodology, and its functionality is demonstrated using two cases. With Meander, users can explore variations at different levels of resolution and simultaneously compare up to four different individuals against a common reference. The application was developed using Java version 1.6 and Processing.org and can be run on any platform. It can be found at http://homes.esat.kuleuven.be/~bioiuser/meander.

Finding the targets of a drug by integration of gene expression data with a protein interaction network.

Polypharmacology, which focuses on designing drugs that bind efficiently to multiple targets, has emerged as a new strategic trend in today's drug discovery research. Many successful drugs achieve their effects via multi-target interactions. However, these targets are largely unknown for both marketed drugs and drugs in development. A better knowledge of a drug's mode of action could be of substantial value to future drug development, in particular for side effect prediction and drug repositioning. We propose a network-based computational method for drug target prediction, applicable on a genome-wide scale. Our approach relies on the analysis of gene expression following drug treatment in the context of a functional protein association network. By diffusing differential expression signals to neighboring or correlated nodes in the network, genes are prioritized as potential targets based on the transcriptional response of functionally related genes. Different diffusion strategies were evaluated on 235 publicly available gene expression datasets for treatment with bioactive molecules having a known target. AUC values of up to more than 90% demonstrate the effectiveness of our approach and indicate the predictive power of integrating experimental gene expression data with prior knowledge from protein association networks.

BMPR1A is a candidate gene for congenital heart defects associated with the recurrent 10q22q23 deletion syndrome.

Congenital heart defects (CHD) are associated with the recurrent 10q22q23 deletion syndrome and with partially overlapping distal 10q23.2.q23.31 microdeletions. We report on a de novo intragenic deletion of the BMPR1A gene in a normally developing adolescent boy with short stature, delayed puberty, facial dysmorphism and an atrioventricular septal defect. Based on this finding, complemented with computational prioritization data and molecular evidence in literature, the critical region for CHD on 10q23 can be downsized to a single gene, BMPR1A. Although loss-of-function mutations in BMPR1A typically result in juvenile polyposis syndrome, none of the patients with the typical 10q22q23 microdeletion syndrome, comprising this gene, were reported to have juvenile polyposis thus far. We reason that, even in the absence of juvenile polyposis syndrome, sequencing and copy number analysis of BMPR1A should be considered in patients with (atrioventricular) septal defects, especially when associated with facial dysmorphism and anomalous growth.

Chronic lymphocytic leukemia and prolymphocytic leukemia with MYC translocations: a subgroup with an aggressive disease course.

Translocations involving MYC are rare in chronic lymphocytic leukemia (CLL), and up to now, their prognostic significance remains unclear. We report the characteristics of 21 patients with CLL and nine patients with prolymphocytic leukemia (PLL), diagnosed in multiple centers (n = 13), which showed an MYC translocation demonstrated by fluorescence in situ hybridization. The prevalence was estimated to be <1%. Advanced age and male predominance were observed. Morphological analysis frequently revealed the presence of prolymphocytes. A typical "CLL-immunophenotype" was found in four of nine cases with PLL. Moreover, CD5 and CD23 were frequently expressed in PLL. The latter findings are atypical for PLL and may suggest transformation or progression of an underlying CLL. MYC translocations were frequently observed with concomitant adverse cytogenetic markers, such as del(11q) (n = 8/30) and/or del(17p)/monosomy 17 (n = 7/30). In addition, the presence of unbalanced translocations (n = 24 in 13/30 cases) and complex karyotype (n = 16/30) were frequent in cases with MYC translocations. Altogether, del(17p)/monosomy 17, del(11q), and/or complex karyotype were observed in 22 of 30 patients. Survival outcome was poor: the median time to treatment was only 5 months, and overall survival (OS) from clinical diagnosis and from genetic detection was 71 and 19 months, respectively. In conclusion, CLL/PLL with MYC translocations is a rare entity, which seems to be associated with adverse prognostic features and unfavorable outcome.

Breakage-fusion-bridge cycles leading to inv dup del occur in human cleavage stage embryos.

Recently, a high incidence of chromosome instability (CIN) was reported in human cleavage stage embryos. Based on the copy number changes that were observed in the blastomeres it was hypothesized that chromosome breakages and fusions occur frequently in cleavage stage human embryos and instigate subsequent breakage-fusion-bridge cycles. In addition, it was hypothesized that the DNA breaks present in spermatozoa could trigger this CIN. To test these hypotheses, we genotyped both parents as well as 93 blastomeres from 24 IVF embryos and developed a novel single nucleotide polymorphism (SNP) array-based algorithm to determine the parental origin of (aberrant) loci in single cells. Paternal as well as maternal alleles were commonly rearranged in the blastomeres indicating that sperm-specific DNA breaks do not explain the majority of these structural variants. The parent-of-origin analyses together with microarray-guided FISH analyses demonstrate the presence of inv dup del chromosomes as well as more complex rearrangements. These data provide unequivocal evidence for breakage-fusion-bridge cycles in those embryos and suggest that the human cleavage stage embryo is a major source of chromosomal disorders.