Designing interpretable deep learning applications for functional genomics: a quantitative analysis.

Deep learning applications have had a profound impact on many scientific fields, including functional genomics. Deep learning models can learn complex interactions between and within omics data; however, interpreting and explaining these models can be challenging. Interpretability is essential not only to help progress our understanding of the biological mechanisms underlying traits and diseases but also for establishing trust in these model's efficacy for healthcare applications. Recognizing this importance, recent years have seen the development of numerous diverse interpretability strategies, making it increasingly difficult to navigate the field. In this review, we present a quantitative analysis of the challenges arising when designing interpretable deep learning solutions in functional genomics. We explore design choices related to the characteristics of genomics data, the neural network architectures applied, and strategies for interpretation. By quantifying the current state of the field with a predefined set of criteria, we find the most frequent solutions, highlight exceptional examples, and identify unexplored opportunities for developing interpretable deep learning models in genomics.

Novel multi-omics deconfounding variational autoencoders can obtain meaningful disease subtyping.

Unsupervised learning, particularly clustering, plays a pivotal role in disease subtyping and patient stratification, especially with the abundance of large-scale multi-omics data. Deep learning models, such as variational autoencoders (VAEs), can enhance clustering algorithms by leveraging inter-individual heterogeneity. However, the impact of confounders-external factors unrelated to the condition, e.g. batch effect or age-on clustering is often overlooked, introducing bias and spurious biological conclusions. In this work, we introduce four novel VAE-based deconfounding frameworks tailored for clustering multi-omics data. These frameworks effectively mitigate confounding effects while preserving genuine biological patterns. The deconfounding strategies employed include (i) removal of latent features correlated with confounders, (ii) a conditional VAE, (iii) adversarial training, and (iv) adding a regularization term to the loss function. Using real-life multi-omics data from The Cancer Genome Atlas, we simulated various confounding effects (linear, nonlinear, categorical, mixed) and assessed model performance across 50 repetitions based on reconstruction error, clustering stability, and deconfounding efficacy. Our results demonstrate that our novel models, particularly the conditional multi-omics VAE (cXVAE), successfully handle simulated confounding effects and recover biologically driven clustering structures. cXVAE accurately identifies patient labels and unveils meaningful pathological associations among cancer types, validating deconfounded representations. Furthermore, our study suggests that some of the proposed strategies, such as adversarial training, prove insufficient in confounder removal. In summary, our study contributes by proposing innovative frameworks for simultaneous multi-omics data integration, dimensionality reduction, and deconfounding in clustering. Benchmarking on open-access data offers guidance to end-users, facilitating meaningful patient stratification for optimized precision medicine.

PortPred: Exploiting deep learning embeddings of amino acid sequences for the identification of transporter proteins and their substrates.

The physiology of every living cell is regulated at some level by transporter proteins which constitute a relevant portion of membrane-bound proteins and are involved in the movement of ions, small and macromolecules across bio-membranes. The importance of transporter proteins is unquestionable. The prediction and study of previously unknown transporters can lead to the discovery of new biological pathways, drugs and treatments. Here we present PortPred, a tool to accurately identify transporter proteins and their substrate starting from the protein amino acid sequence. PortPred successfully combines pre-trained deep learning-based protein embeddings and machine learning classification approaches and outperforms other state-of-the-art methods. In addition, we present a comparison of the most promising protein sequence embeddings (Unirep, SeqVec, ProteinBERT, ESM-1b) and their performances for this specific task.

Systemic immune activation profiles in streptococcal necrotizing soft tissue infections: A prospective multicenter study.

OBJECTIVE: Early stages with streptococcal necrotizing soft tissue infections (NSTIs) are often difficult to discern from cellulitis. Increased insight into inflammatory responses in streptococcal disease may guide correct interventions and discovery of novel diagnostic targets. METHODS: Plasma levels of 37 mediators, leucocytes and CRP from 102 patients with ß-hemolytic streptococcal NSTI derived from a prospective Scandinavian multicentre study were compared to those of 23 cases of streptococcal cellulitis. Hierarchical cluster analyses were also performed. RESULTS: Differences in mediator levels between NSTI and cellulitis cases were revealed, in particular for IL-1ß, TNFα and CXCL8 (AUC >0.90). Across streptococcal NSTI etiologies, eight biomarkers separated cases with septic shock from those without, and four mediators predicted a severe outcome. CONCLUSION: Several inflammatory mediators and wider profiles were identified as potential biomarkers of NSTI. Associations of biomarker levels to type of infection and outcomes may be utilized to improve patient care and outcomes.

Association of Plasma Metabolites and Lipoproteins with Rh and ABO Blood Systems in Healthy Subjects.

This study investigated the associations between the levels of 27 plasma metabolites, 114 lipoprotein parameters, determined using nuclear magnetic resonance spectroscopy, and the ABO blood groups and the Rhesus (Rh) blood system in a cohort of n = 840 Italian healthy blood donors of both sexes. We observed good multivariate discrimination between the metabolomic and lipoproteomic profiles of subjects with positive and negative Rh. In contrast, we did not observe significant discrimination for the ABO blood group pairwise comparisons, suggesting only slight metabolic differences between these group-specific metabolic profiles. We report univariate associations (P-value < 0.05) between the subfraction HDL1 related to Apo A1, the subfraction HDL2 related to cholesterol and phospholipids, and the particle number of LDL2 related to free cholesterol, cholesterol, phospholipids, and Apo B and the ABO blood groups; we observed association of the lipid main fraction LDL4 related to free cholesterol, triglycerides, and Apo B; creatine; the particle number of LDL5; the subfraction LDL5 related to Apo B; the particle number of LDL4; and the subfraction LDL4 related to Apo B with Rh blood factors. These results suggest blood group-dependent (re)shaping of lipoprotein metabolism in healthy subjects, which may provide relevant information to explain the differential susceptibility to certain diseases observed in different blood groups.

Analysis of host-pathogen gene association networks reveals patient-specific response to streptococcal and polymicrobial necrotising soft tissue infections.

BACKGROUND: Necrotising soft tissue infections (NSTIs) are rapidly progressing bacterial infections usually caused by either several pathogens in unison (polymicrobial infections) or Streptococcus pyogenes (mono-microbial infection). These infections are rare and are associated with high mortality rates. However, the underlying pathogenic mechanisms in this heterogeneous group remain elusive. METHODS: In this study, we built interactomes at both the population and individual levels consisting of host-pathogen interactions inferred from dual RNA-Seq gene transcriptomic profiles of the biopsies from NSTI patients. RESULTS: NSTI type-specific responses in the host were uncovered. The S. pyogenes mono-microbial subnetwork was enriched with host genes annotated with involved in cytokine production and regulation of response to stress. The polymicrobial network consisted of several significant associations between different species (S. pyogenes, Porphyromonas asaccharolytica and Escherichia coli) and host genes. The host genes associated with S. pyogenes in this subnetwork were characterised by cellular response to cytokines. We further found several virulence factors including hyaluronan synthase, Sic1, Isp, SagF, SagG, ScfAB-operon, Fba and genes upstream and downstream of EndoS along with bacterial housekeeping genes interacting with the human stress and immune response in various subnetworks between host and pathogen. CONCLUSIONS: At the population level, we found aetiology-dependent responses showing the potential modes of entry and immune evasion strategies employed by S. pyogenes, congruent with general cellular processes such as differentiation and proliferation. After stratifying the patients based on the subject-specific networks to study the patient-specific response, we observed different patient groups with different collagens, cytoskeleton and actin monomers in association with virulence factors, immunogenic proteins and housekeeping genes which we utilised to postulate differing modes of entry and immune evasion for different bacteria in relationship to the patients' phenotype.

Evaluation of Single Sample Network Inference Methods for Metabolomics-Based Systems Medicine.

Networks and network analyses are fundamental tools of systems biology. Networks are built by inferring pair-wise relationships among biological entities from a large number of samples such that subject-specific information is lost. The possibility of constructing these sample (individual)-specific networks from single molecular profiles might offer new insights in systems and personalized medicine and as a consequence is attracting more and more research interest. In this study, we evaluated and compared LIONESS (Linear Interpolation to Obtain Network Estimates for Single Samples) and ssPCC (single sample network based on Pearson correlation) in the metabolomics context of metabolite-metabolite association networks. We illustrated and explored the characteristics of these two methods on (i) simulated data, (ii) data generated from a dynamic metabolic model to simulate real-life observed metabolite concentration profiles, and (iii) 22 metabolomic data sets and (iv) we applied single sample network inference to a study case pertaining to the investigation of necrotizing soft tissue infections to show how these methods can be applied in metabolomics. We also proposed some adaptations of the methods that can be used for data exploration. Overall, despite some limitations, we found single sample networks to be a promising tool for the analysis of metabolomics data.

Differential Network Analysis Reveals Molecular Determinants Associated with Blood Pressure and Heart Rate in Healthy Subjects.

There is mounting evidence that subclinical nonpathological high blood pressure and heart rate during youth and adulthood steadily increase the risk of developing a cardiovascular disease at a later stage. For this reason, it is important to understand the mechanisms underlying the subclinical elevation of blood pressure and heart rate in healthy, relatively young individuals. In the present study, we present a network-based metabolomic study of blood plasma metabolites and lipids measured using nuclear magnetic resonance spectroscopy on 841 adult healthy blood donor volunteers, which were stratified for subclinical low and high blood pressure (systolic and diastolic) and heart rate. Our results indicate a rewiring of metabolic pathways active in high and low groups, indicating that the subjects with subclinical high blood pressure and heart rate could present latent cardiometabolic dysregulations.

Analysis of Metabolite and Lipid Association Networks Reveals Molecular Mechanisms Associated with 3-Month Mortality and Poor Functional Outcomes in Patients with Acute Ischemic Stroke after Thrombolytic Treatment with Recombinant Tissue Plasminogen Activator.

Here, we present an integrated multivariate, univariate, network reconstruction and differential analysis of metabolite-metabolite and metabolite-lipid association networks built from an array of 18 serum metabolites and 110 lipids identified and quantified through nuclear magnetic resonance spectroscopy in a cohort of 248 patients, of which 22 died and 82 developed a poor functional outcome within 3 months from acute ischemic stroke (AIS) treated with intravenous recombinant tissue plasminogen activator. We explored differences in metabolite and lipid connectivity of patients who did not develop a poor outcome and who survived the ischemic stroke from the related opposite conditions. We report statistically significant differences in the connectivity patterns of both low- and high-molecular-weight metabolites, implying underlying variations in the metabolic pathway involving leucine, glycine, glutamine, tyrosine, phenylalanine, citric, lactic, and acetic acids, ketone bodies, and different lipids, thus characterizing patients' outcomes. Our results evidence the promising and powerful role of the metabolite-metabolite and metabolite-lipid association networks in investigating molecular mechanisms underlying AIS patient's outcome.

Phenotype and multi-omics comparison of Staphylococcus and Streptococcus uncovers pathogenic traits and predicts zoonotic potential.

BACKGROUND: Staphylococcus and Streptococcus species can cause many different diseases, ranging from mild skin infections to life-threatening necrotizing fasciitis. Both genera consist of commensal species that colonize the skin and nose of humans and animals, and of which some can display a pathogenic phenotype. RESULTS: We compared 235 Staphylococcus and 315 Streptococcus genomes based on their protein domain content. We show the relationships between protein persistence and essentiality by integrating essentiality predictions from two metabolic models and essentiality measurements from six large-scale transposon mutagenesis experiments. We identified clusters of strains within species based on proteins associated to similar biological processes. We built Random Forest classifiers that predicted the zoonotic potential. Furthermore, we identified shared attributes between of Staphylococcus aureus and Streptococcus pyogenes that allow them to cause necrotizing fasciitis. CONCLUSIONS: Differences observed in clustering of strains based on functional groups of proteins correlate with phenotypes such as host tropism, capability to infect multiple hosts and drug resistance. Our method provides a solid basis towards large-scale prediction of phenotypes based on genomic information.

In-Pero: Exploiting Deep Learning Embeddings of Protein Sequences to Predict the Localisation of Peroxisomal Proteins.

Peroxisomes are ubiquitous membrane-bound organelles, and aberrant localisation of peroxisomal proteins contributes to the pathogenesis of several disorders. Many computational methods focus on assigning protein sequences to subcellular compartments, but there are no specific tools tailored for the sub-localisation (matrix vs. membrane) of peroxisome proteins. We present here In-Pero, a new method for predicting protein sub-peroxisomal cellular localisation. In-Pero combines standard machine learning approaches with recently proposed multi-dimensional deep-learning representations of the protein amino-acid sequence. It showed a classification accuracy above 0.9 in predicting peroxisomal matrix and membrane proteins. The method is trained and tested using a double cross-validation approach on a curated data set comprising 160 peroxisomal proteins with experimental evidence for sub-peroxisomal localisation. We further show that the proposed approach can be easily adapted (In-Mito) to the prediction of mitochondrial protein localisation obtaining performances for certain classes of proteins (matrix and inner-membrane) superior to existing tools.

Metabolomics of Milk Reflects a Negative Energy Balance in Cows.

Dairy cows can experience a negative energy balance (NEB) in early lactation when feed intake is too low to meet the energy requirements for body maintenance and milk production. Metabolic changes occur in mammary gland cells of animals experiencing a negative energy balance. We studied these metabolic changes in milk samples from dairy cows in relation to energy balance status using liquid chromatography-mass spectrometry (QQQ-LC-MS) and nuclear magnetic resonance (1H NMR). NMR and LC-MS techniques are complementary techniques that enabled a comprehensive overview of milk metabolites in our study. Energy balance and milk samples were obtained from 87 dairy cows. A total of 55 milk metabolites were reliably detected, of which 15 metabolites were positively correlated to energy balance and 20 were negatively correlated to energy balance. Cows in NEB produced more milk with increased milk fat yield and higher concentrations of citrate, cis-aconitate, creatinine, glycine, phosphocreatine, galactose-1-phosphate, glucose-1-phosphate, UDP-N-acetyl-galactosamine, UDP-N-acetyl-glucosamine, and phosphocholine but lower concentrations of choline, ethanolamine, fucose, N-acetyl-neuraminic acid, N-acetyl-glucosamine, and N-acetyl-galactosamine. During NEB, we observed an increased leakage of cellular content, increased synthesis of nucleic acids and cell membrane phospholipids, an increase in one-carbon metabolic processes, and an increase in lipid-triglyceride anabolism. Overall, both apoptosis combined with cellular renewal is paramount in the mammary gland in cows in NEB.

Integrated Univariate, Multivariate, and Correlation-Based Network Analyses Reveal Metabolite-Specific Effects on Bacterial Growth and Biofilm Formation in Necrotizing Soft Tissue Infections.

Necrotizing soft-tissue infections (NSTIs) have multiple causes, risk factors, anatomical locations, and pathogenic mechanisms. In patients with NSTI, circulating metabolites may serve as a substrate having impact on bacterial adaptation at the site of infection. Metabolic signatures associated with NSTI may reveal the potential to be useful as diagnostic and prognostic markers and novel targets for therapy. This study used untargeted metabolomics analyses of plasma from NSTI patients (n = 34) and healthy (noninfected) controls (n = 24) to identify the metabolic signatures and connectivity patterns among metabolites associated with NSTI. Metabolite-metabolite association networks were employed to compare the metabolic profiles of NSTI patients and noninfected surgical controls. Out of 97 metabolites detected, the abundance of 33 was significantly altered in NSTI patients. Analysis of metabolite-metabolite association networks showed a more densely connected network: specifically, 20 metabolites differentially connected between NSTI and controls. A selected set of significantly altered metabolites was tested in vitro to investigate potential influence on NSTI group A streptococcal strain growth and biofilm formation. Using chemically defined media supplemented with the selected metabolites, ornithine, ribose, urea, and glucuronic acid, revealed metabolite-specific effects on both bacterial growth and biofilm formation. This study identifies for the first time an NSTI-specific metabolic signature with implications for optimized diagnostics and therapies.

Differential Network Analysis Reveals Metabolic Determinants Associated with Mortality in Acute Myocardial Infarction Patients and Suggests Potential Mechanisms Underlying Different Clinical Scores Used To Predict Death.

We present here the differential analysis of metabolite-metabolite association networks constructed from an array of 24 serum metabolites identified and quantified via nuclear magnetic resonance spectroscopy in a cohort of 825 patients of which 123 died within 2 years from acute myocardial infarction (AMI). We investigated differences in metabolite connectivity of patients who survived, at 2 years, the AMI event, and we characterized metabolite-metabolite association networks specific to high and low risks of death according to four different risk parameters, namely, acute coronary syndrome classification, Killip, Global Registry of Acute Coronary Events risk score, and metabolomics NOESY RF risk score. We show significant differences in the connectivity patterns of several low-molecular-weight molecules, implying variations in the regulation of several metabolic pathways regarding branched-chain amino acids, alanine, creatinine, mannose, ketone bodies, and energetic metabolism. Our results demonstrate that the characterization of metabolite-metabolite association networks is a promising and powerful tool to investigate AMI patients according to their outcomes at a molecular level.

Systems Biology and Biomarkers in Necrotizing Soft Tissue Infections.

In necrotizing soft tissue infection (NSTI) there is a need to identify biomarker sets that can be used for diagnosis and disease management. The INFECT study was designed to obtain such insights through the integration of patient data and results from different clinically relevant experimental models by use of systems biology approaches. This chapter describes the current state of biomarkers in NSTI and how biomarkers are categorized. We introduce the fundamentals of top-down systems biology approaches including analysis tools and we review the use of current methods and systems biology approaches to biomarker discover. Further, we discuss how different "omics" signatures (gene expression, protein, and metabolites) from NSTI patient samples can be used to identify key host and pathogen factors involved in the onset and development of infection, as well as exploring associations to disease outcomes.

Systems and Precision Medicine in Necrotizing Soft Tissue Infections.

Necrotizing soft tissue infections (NSTI) are multifactorial and characterized by dysfunctional, time dependent, highly varying hyper- to hypo-inflammatory host responses contributing to disease severity. Furthermore, host-pathogen interactions are diverse and difficult to identify and characterize, due to the many different disease endotypes. There is a need for both refined bedside diagnostics as well as novel targeted treatment options to improve outcome in NSTI. In order to achieve clinically relevant results and to guide preclinical and clinical research the vast amount of fragmented clinical and experimental datasets, which often include omics data at different levels (transcriptomics, proteomics, metabolomics, etc.), need to be organized, harmonized, integrated, and analyzed taking into account the Big Data nature of these datasets. In this chapter, we address these matters from a systems perspective and yet personalized approach. The chapter provides an overview on the increasingly more frequent use of Big Data and Artificial Intelligence (AI) to aggregate and generate knowledge from burgeoning clinical and biochemical information, addresses the challenges to manage this information, and summarizes current efforts to develop robust computer-aided clinical decision support systems so to tackle the serious challenges in NSTI diagnosis, stratification, and optimized tailored therapy.

Short communication: Prediction of hyperketonemia in dairy cows in early lactation using on-farm cow data and net energy intake by partial least square discriminant analysis.

The objectives of this study were (1) to evaluate if hyperketonemia in dairy cows (defined as plasma ß-hydroxybutyrate ≥1.0 mmol/L) can be predicted using on-farm cow data either in current or previous lactation week, and (2) to study if adding individual net energy intake (NEI) can improve the predictive ability of the model. Plasma ß-hydroxybutyrate concentration, on-farm cow data (milk yield, percentage of fat, protein and lactose, fat- and protein-corrected milk yield, body weight, body weight change, dry period length, parity, and somatic cell count), and NEI of 424 individual cows were available weekly through lactation wk 1 to 5 postpartum. To predict hyperketonemia in dairy cows, models were first trained by partial least square discriminant analysis, using on-farm cow data in the same or previous lactation week. Second, NEI was included in models to evaluate the improvement of the predictability of the models. Through leave-one trial-out cross-validation, models were evaluated by accuracy (the ratio of the sum of true positive and true negative), sensitivity (68.2% to 84.9%), specificity (61.5% to 98.7%), positive predictive value (57.7% to 98.7%), and negative predictive value (66.2% to 86.1%) to predict hyperketonemia of dairy cows. Through lactation wk 1 to 5, the accuracy to predict hyperketonemia using data in the same week was 64.4% to 85.5% (on-farm cow data only), 66.1% to 87.0% (model including NEI), and using data in the previous week was 58.5% to 82.0% (on-farm cow data only), 59.7% to 85.1% (model including NEI). An improvement of the accuracy of the model due to including NEI ranged among lactation weeks from 1.0% to 4.4% when using data in the same lactation week and 0.2% to 6.6% when using data in the previous lactation week. In conclusion, trained models via partial least square discriminant analysis have potential to predict hyperketonemia in dairy cows not only using data in the current lactation week, but also using data in the previous lactation week. Net energy intake can improve the accuracy of the model, but only to a limited extent. Besides NEI, body weight, body weight change, milk fat, and protein content were important variables to predict hyperketonemia, but their rank of importance differed across lactation weeks.

Relationship between energy balance and metabolic profiles in plasma and milk of dairy cows in early lactation.

Negative energy balance in dairy cows in early lactation is related to alteration of metabolic status. However, the relationships among energy balance, metabolic profile in plasma, and metabolic profile in milk have not been reported. In this study our aims were: (1) to reveal the metabolic profiles of plasma and milk by integrating results from nuclear magnetic resonance (NMR) with data from liquid chromatography triple quadrupole mass spectrometry (LC-MS); and (2) to investigate the relationship between energy balance and the metabolic profiles of plasma and milk. For this study 24 individual dairy cows (parity 2.5 ± 0.5; mean ± standard deviation) were studied in lactation wk 2. Body weight (mean ± standard deviation; 627.4 ± 56.4 kg) and milk yield (28.1 ± 6.7 kg/d; mean ± standard deviation) were monitored daily. Milk composition (fat, protein, and lactose) and net energy balance were calculated. Plasma and milk samples were collected and analyzed using LC-MS and NMR. From all plasma metabolites measured, 27 were correlated with energy balance. These plasma metabolites were related to body reserve mobilization from body fat, muscle, and bone; increased blood flow; and gluconeogenesis. From all milk metabolites measured, 30 were correlated with energy balance. These milk metabolites were related to cell apoptosis and cell proliferation. Nine metabolites detected in both plasma and milk were correlated with each other and with energy balance. These metabolites were mainly related to hyperketonemia; ß-oxidation of fatty acids; and one-carbon metabolism. The metabolic profiles of plasma and milk provide an in-depth insight into the physiological pathways of dairy cows in negative energy balance in early lactation. In addition to the classical indicators for energy balance (e.g., ß-hydroxybutyrate, acetone, and glucose), the current study presents some new metabolites (e.g., glycine in plasma and milk; kynurenine, panthothenate, or arginine in plasma) in lactating dairy cows that are related to energy balance and may be of interest as new indicators for energy balance.

Simulation and Reconstruction of Metabolite-Metabolite Association Networks Using a Metabolic Dynamic Model and Correlation Based Algorithms.

Biological networks play a paramount role in our understanding of complex biological phenomena, and metabolite-metabolite association networks are now commonly used in metabolomics applications. In this study we evaluate the performance of several network inference algorithms (PCLRC, MRNET, GENIE3, TIGRESS, and modifications of the MRNET algorithm, together with standard Pearson's and Spearman's correlation) using as a test case data generated using a dynamic metabolic model describing the metabolism of arachidonic acid (consisting of 83 metabolites and 131 reactions) and simulation individual metabolic profiles of 550 subjects. The quality of the reconstructed metabolite-metabolite association networks was assessed against the original metabolic network taking into account different degrees of association among the metabolites and different sample sizes and noise levels. We found that inference algorithms based on resampling and bootstrapping perform better when correlations are used as indexes to measure the strength of metabolite-metabolite associations. We also advocate for the use of data generated using dynamic models to test the performance of algorithms for network inference since they produce correlation patterns that are more similar to those observed in real metabolomics data.

SAPP: functional genome annotation and analysis through a semantic framework using FAIR principles.

Summary: To unlock the full potential of genome data and to enhance data interoperability and reusability of genome annotations we have developed SAPP, a Semantic Annotation Platform with Provenance. SAPP is designed as an infrastructure supporting FAIR de novo computational genomics but can also be used to process and analyze existing genome annotations. SAPP automatically predicts, tracks and stores structural and functional annotations and associated dataset- and element-wise provenance in a Linked Data format, thereby enabling information mining and retrieval with Semantic Web technologies. This greatly reduces the administrative burden of handling multiple analysis tools and versions thereof and facilitates multi-level large scale comparative analysis. Availability and implementation: SAPP is written in JAVA and freely available at https://gitlab.com/sapp and runs on Unix-like operating systems. The documentation, examples and a tutorial are available at https://sapp.gitlab.io. Contact: jasperkoehorst@gmail.com or peter.schaap@wur.nl.