Composite Graph Neural Networks for Molecular Property Prediction.

Graph Neural Networks have proven to be very valuable models for the solution of a wide variety of problems on molecular graphs, as well as in many other research fields involving graph-structured data. Molecules are heterogeneous graphs composed of atoms of different species. Composite graph neural networks process heterogeneous graphs with multiple-state-updating networks, each one dedicated to a particular node type. This approach allows for the extraction of information from s graph more efficiently than standard graph neural networks that distinguish node types through a one-hot encoded type of vector. We carried out extensive experimentation on eight molecular graph datasets and on a large number of both classification and regression tasks. The results we obtained clearly show that composite graph neural networks are far more efficient in this setting than standard graph neural networks.

Protein-Protein Interfaces: A Graph Neural Network Approach.

Protein-protein interactions (PPIs) are fundamental processes governing cellular functions, crucial for understanding biological systems at the molecular level. Compared to experimental methods for PPI prediction and site identification, computational deep learning approaches represent an affordable and efficient solution to tackle these problems. Since protein structure can be summarized as a graph, graph neural networks (GNNs) represent the ideal deep learning architecture for the task. In this work, PPI prediction is modeled as a node-focused binary classification task using a GNN to determine whether a generic residue is part of the interface. Biological data were obtained from the Protein Data Bank in Europe (PDBe), leveraging the Protein Interfaces, Surfaces, and Assemblies (PISA) service. To gain a deeper understanding of how proteins interact, the data obtained from PISA were assembled into three datasets: Whole, Interface, and Chain, consisting of data on the whole protein, couples of interacting chains, and single chains, respectively. These three datasets correspond to three different nuances of the problem: identifying interfaces between protein complexes, between chains of the same protein, and interface regions in general. The results indicate that GNNs are capable of solving each of the three tasks with very good performance levels.

Postintubation Sedation Practices Within Multiple Emergency Departments Across a Large Pediatric Health Care Organization.

OBJECTIVE: Children requiring rapid or standard sequence intubation are at risk of experiencing paralysis without adequate sedation when the duration of neuromuscular blockade exceeds the duration of sedation provided by the induction agent. The objective of this study was to evaluate the rate of appropriately timed postintubation sedation (PIS; defined as the administration of PIS before the clinical effects of the induction agent have dissipated) in patients requiring intubation across multiple emergency department/urgent care sites within a large pediatric health care organization. METHODS: This retrospective cohort study included patients admitted to 1 of 6 affiliated pediatric emergency department or urgent care sites who were intubated with an induction agent and neuromuscular blocker between January 2016 and December 2021. Patients were excluded if they were intubated in the setting of status epilepticus or cardiac arrest. Stepwise logistic regression identified factors associated with appropriately timed PIS. RESULTS: A total of 283 patients met the inclusion criteria (mean age, 8 ± 7.6 years; 56% male). Two hundred thirty-eight patients (83%) received some form of PIS (105 [37%] received appropriately timed PIS and 133 [47%] received delayed PIS), and 45 patients (16%) received no PIS. The median time to receive PIS following administration of the induction agent was 21 minutes (interquartile range, 11-40 minutes). Patients induced with fentanyl were the least likely to receive PIS, whereas patients induced with etomidate were the most likely. However, because of the short duration of etomidate, most patients induced with etomidate failed to receive PIS in a timely manner. CONCLUSIONS: Delayed PIS is common and may result in periods of ongoing paralysis without adequate sedation. Emergency department providers and pharmacists must recognize the brevity of some induction agents and provide more timely PIS.

Cost-Effectiveness Analysis of New Beta-Lactam Beta-Lactamase Inhibitor Antibiotics Versus Colistin for the Treatment of Carbapenem-Resistant Infections.

Introduction: Carbapenem-resistant organisms (CROs) present a serious public health problem. Limited treatment options has led to increased use of colistin and polymyxin. Since 2014, the US Food and Drug Administration approved 4 new beta-lactam beta-lactamase inhibitor (BLBLI) combination antibiotics with activity against CROs. These new antibiotics have been shown to be more effective and less toxic than colistin and polymyxin but are considerably more expensive. This study evaluated the cost-effectiveness of the new BLBLIs versus colistin-based therapy for the treatment of CROs. Methods: A decision-tree microsimulation model was used to evaluate the cost effectiveness of the new BLBLIs versus colistin-based therapy for the treatment of CROs. Treatment groups differed in risk of mortality and risk of an acute kidney injury (AKI). The relative risk of mortality was determined by creating a meta-analysis comparing new BLBLIs to colistin. Cost inputs included medication costs and the cost to treat an AKI. The primary outcomes include quality-adjusted life years (QALYs) and incremental cost-effectiveness ratio (ICER). Model inputs included: clinical outcomes and adverse events (30-day mortality and AKI); cost of treatment and adverse drug events; and health utilities. A 3% discount was applied for outcomes. A lifetime horizon was used from the perspective of the US healthcare system with a willingness-to-pay (WTP) threshold of $100 000. A sensitivity analysis was done to incorporate uncertainty. Results: The meta-analysis found the treatment with a new BLBLI was associated with a 50% decrease in the relative risk of 30-day mortality compared to colistin (RR 0.47, 95% CI 0.25-0.88). Treatment with a new BLBLI cost $16 200 and produced 11.5 QALYs, on average. The average colistin based regimen cost $3500 and produced 8.3 QALYs. The new BLBLIs were determined to be cost-effective with an ICER of $3900 per QALY gained. Treatment with a BLBLI remained cost-effective under all uncertainty scenarios tested. Conclusion: New BLBLIs are cost-effective compared to colistin for the treatment of CROs and are associated with improved mortality and fewer AKI events. The use of colistin should be reserved for cases where new BLBLIs are not available or there is documented resistance to these new antibiotics.

Discharge Delays and Costs Associated With Outpatient Parenteral Antimicrobial Therapy for High-Priced Antibiotics.

BACKGROUND: Outpatient parenteral antimicrobial therapy (OPAT) is a widely used, safe, and cost-effective treatment. Most public and private insurance providers require prior authorization (PA) for OPAT, yet the impact of the inpatient PA process is not known. Our aim was to characterize discharge barriers and PA delays associated with high-priced OPAT antibiotics. METHODS: This was an institutional review board-approved study of adult patients discharged with daptomycin, ceftaroline, ertapenem, and novel beta-lactam-beta-lactamase inhibitor combinations from January 2017 to December 2017. Patients with an OPAT PA delay were compared with patients without a delay. The primary endpoint was total direct hospital costs from the start of treatment. RESULTS: Two-hundred patients were included: 141 (71%) no OPAT delay vs 59 (30%) OPAT delay. More patients with a PA delay were discharged to a subacute care facility compared with an outpatient setting: 37 (63%) vs 52 (37%), P = .001. Discharge delays and median total direct hospital costs were higher for patients with OPAT delays: 31 (53%) vs 21 (15%), P < .001 and $19 576 (interquartile range [IQR], 10 056-37 038) vs $7770 (IQR, 3031-13 974), P < .001. In multiple variable regression, discharge to a subacute care facility was associated with an increased odds of discharge delay, age >64 years was associated with a decreased odds of discharge delay. CONCLUSIONS: OPAT with high-priced antibiotics requires significant care coordination. PA delays are common and contribute to discharge delays. OPAT transitions of care represent an opportunity to improve patient care and address access barriers.

A possible strategy to fight COVID-19: Interfering with spike glycoprotein trimerization.

The recent release of COVID-19 spike glycoprotein allows detailed analysis of the structural features that are required for stabilizing the infective form of its quaternary assembly. Trying to disassemble the trimeric structure of COVID-19 spike glycoprotein, we analyzed single protomer surfaces searching for concave moieties that are located at the three protomer-protomer interfaces. The presence of some druggable pockets at these interfaces suggested that some of the available drugs in Drug Bank could destabilize the quaternary spike glycoprotein formation by binding to these pockets, therefore interfering with COVID-19 life cycle. The approach we propose here can be an additional strategy to fight against the deadly virus. Ligands of COVID-19 spike glycoprotein that we have predicted in the present computational investigation, might be the basis for new experimental studies in vitro and in vivo.

Clinical characteristics and predictors of survival in adults with coronavirus disease 2019 receiving tocilizumab.

Coronavirus disease 2019 (COVID-19) can progress to cytokine storm that is associated with organ dysfunction and death. The purpose of the present study is to determine clinical characteristics associated with 28 day in-hospital survival in patients with coronavirus disease 2019 (COVID-19) that received tocilizumab. This was a retrospective observational cohort study conducted at a five hospital health system in Michigan, United States. Adult patients with confirmed COVID-19 that were admitted to the hospital and received tocilizumab for cytokine storm from March 1, 2020 through April 3, 2020 were included. Patients were grouped into survivors and non-survivors based on 28 day in-hospital mortality. Study day 0 was defined as the day tocilizumab was administered. Factors independently associated with in-hospital survival at 28 days after tocilizumab administration were assessed. Epidemiologic, demographic, laboratory, prognostic scores, treatment, and outcome data were collected and analyzed. Clinical response was collected and defined as a decline of two levels on a six-point ordinal scale of clinical status or discharged alive from the hospital. Of the 81 patients included, the median age was 64 (58-71) years and 56 (69.1%) were male. The 28 day in-hospital mortality was 43.2%. There were 46 (56.8%) patients in the survivors and 35 (43.2%) in the non-survivors group. On study day 0 no differences were noted in demographics, clinical characteristics, severity of illness scores, or treatments received between survivors and non-survivors. C-reactive protein was significantly higher in the non-survivors compared to survivors. Compared to non-survivors, recipients of tocilizumab within 12 days of symptom onset was independently associated with survival (adjusted OR: 0.296, 95% CI: 0.098-0.889). SOFA score ≥8 on day 0 was independently associated with mortality (adjusted OR: 2.842, 95% CI: 1.042-7.753). Clinical response occurred more commonly in survivors than non-survivors (80.4% vs. 5.7%; p < 0.001). Improvements in the six-point ordinal scale and SOFA score were observed in survivors after tocilizumab. Early receipt of tocilizumab in patients with severe COVID-19 was an independent predictor for in-hospital survival at 28 days.

A topological description of loss surfaces based on Betti Numbers.

In the context of deep learning models, attention has recently been paid to studying the surface of the loss function in order to better understand training with methods based on gradient descent. This search for an appropriate description, both analytical and topological, has led to numerous efforts in identifying spurious minima and characterize gradient dynamics. Our work aims to contribute to this field by providing a topological measure for evaluating loss complexity in the case of multilayer neural networks. We compare deep and shallow architectures with common sigmoidal activation functions by deriving upper and lower bounds for the complexity of their respective loss functions and revealing how that complexity is influenced by the number of hidden units, training models, and the activation function used. Additionally, we found that certain variations in the loss function or model architecture, such as adding an ℓ2 regularization term or implementing skip connections in a feedforward network, do not affect loss topology in specific cases.

A Mobile App for Detecting Potato Crop Diseases.

Artificial intelligence techniques are now widely used in various agricultural applications, including the detection of devastating diseases such as late blight (Phytophthora infestans) and early blight (Alternaria solani) affecting potato (Solanum tuberorsum L.) crops. In this paper, we present a mobile application for detecting potato crop diseases based on deep neural networks. The images were taken from the PlantVillage dataset with a batch of 1000 images for each of the three identified classes (healthy, early blight-diseased, late blight-diseased). An exploratory analysis of the architectures used for early and late blight diagnosis in potatoes was performed, achieving an accuracy of 98.7%, with MobileNetv2. Based on the results obtained, an offline mobile application was developed, supported on devices with Android 4.1 or later, also featuring an information section on the 27 diseases affecting potato crops and a gallery of symptoms. For future work, segmentation techniques will be used to highlight the damaged region in the potato leaf by evaluating its extent and possibly identifying different types of diseases affecting the same plant.

A Deep Learning Approach to the Prediction of Drug Side-Effects on Molecular Graphs.

Predicting drug side effects before they occur is a critical task for keeping the number of drug-related hospitalizations low and for improving drug discovery processes. Automatic predictors of side-effects generally are not able to process the structure of the drug, resulting in a loss of information. Graph neural networks have seen great success in recent years, thanks to their ability of exploiting the information conveyed by the graph structure and labels. These models have been used in a wide variety of biological applications, among which the prediction of drug side-effects on a large knowledge graph. Exploiting the molecular graph encoding the structure of the drug represents a novel approach, in which the problem is formulated as a multi-class multi-label graph-focused classification. We developed a methodology to carry out this task, using recurrent Graph Neural Networks, and building a dataset from freely accessible and well established data sources. The results show that our method has an improved classification capability, under many parameters and metrics, with respect to previously available predictors. The method is not ready for clinical tests yet, as the specificity is still below the preliminary 25% threshold. Future efforts will aim at improving this aspect.

Risk Factors Associated With Opioid/Benzodiazepine Iatrogenic Withdrawal Syndrome in COVID-19 Acute Respiratory Distress Syndrome.

BackgroundMechanically ventilated COVID-19 acute respiratory distress syndrome (ARDS) patients often receive deeper sedation and analgesia to maintain respiratory compliance and minimize staff exposure, which incurs greater risk of iatrogenic withdrawal syndrome (IWS) and has been associated with worse patient outcomes. Objective: To identify potential risk factors and differences in patient outcomes associated with the development of IWS in COVID-19 ARDS patients. Methods: Retrospective analysis of ventilated COVID-19 ARDS intensive care unit (ICU) patients who received continuous intravenous (IV) analgesia and sedation for ≥5 days from March 2020-May 2021. Patients were classified as IWS and non-IWS based on receipt of scheduled oral sedative/analgesic regimens after cessation of IV therapy. Risk factors were assessed in univariate analyses and multivariable modeling. Results: A total of 115 patients were included. The final multivariable model showed: (1) each additional day of IV opioid therapy was associated with an 8% increase in odds of IWS (95% CI, 1.02-1.14), (2) among sedatives, receipt of lorazepam was associated with 3 times higher odds of IWS (95% CI 1.12-8.15), and (3) each 1-point increase in Simplified Acute Physiology Score (SAPS) II was associated with a 4% reduction in odds of IWS (95% CI 0.93-0.999). Conclusion: Prolonged and high dose exposures to IV opioids and benzodiazepines should be limited when possible. Additional prospective studies are needed to identify modifiable risk factors to prevent IWS.

Modular Multi-Source Prediction of Drug Side-Effects With DruGNN.

Drug Side-Effects (DSEs) have a high impact on public health, care system costs, and drug discovery processes. Predicting the probability of side-effects, before their occurrence, is fundamental to reduce this impact, in particular on drug discovery. Candidate molecules could be screened before undergoing clinical trials, reducing the costs in time, money, and health of the participants. Drug side-effects are triggered by complex biological processes involving many different entities, from drug structures to protein-protein interactions. To predict their occurrence, it is necessary to integrate data from heterogeneous sources. In this work, such heterogeneous data is integrated into a graph dataset, expressively representing the relational information between different entities, such as drug molecules and genes. The relational nature of the dataset represents an important novelty for drug side-effect predictors. Graph Neural Networks (GNNs) are exploited to predict DSEs on our dataset with very promising results. GNNs are deep learning models that can process graph-structured data, with minimal information loss, and have been applied on a wide variety of biological tasks. Our experimental results confirm the advantage of using relationships between data entities, suggesting interesting future developments in this scope. The experimentation also shows the importance of specific subsets of data in determining associations between drugs and side-effects.

Patient-Level Exposure to Actionable Pharmacogenomic Medications in a Nationally Representative Insurance Claims Database.

BACKGROUND: The prevalence of exposure to pharmacogenomic medications is well established but little is known about how long patients are exposed to these medications. AIM: Our objective was to describe the amount of exposure to actionable pharmacogenomic medications using patient-level measures among a large nationally representative population using an insurance claims database. METHODS: Our retrospective cohort study included adults (18+ years) from the IQVIA PharMetrics® Plus for Academics claims database with incident fills of 72 Clinical Pharmacogenetics Implementation Consortium level A, A/B, or B medications from January 2012 through September 2018. Patient-level outcomes included the proportion of days covered (PDC), number of fills, and average days supplied per fill over a 12-month period. RESULTS: Over 1 million fills of pharmacogenetic medications were identified for 605,355 unique patients. The mean PDC for all medications was 0.21 (SD 0.3), suggesting patients were exposed 21% (77 days) of the year. Medications with the highest PDC (0.55-0.89) included ivacaftor, tamoxifen, clopidogrel, HIV medications, transplant medications, and statins; with the exception of statins, these medications were initiated by fewer patients. Pharmacogenomic medications were filled an average of 2.8 times (SD 3.0, range 1-81) during the year following the medication's initiation, and the average days supplied for each fill was 22.3 days (SD 22.4, range 1-180 days). CONCLUSION: Patient characteristics associated with more medication exposure were male sex, older age, and comorbid chronic conditions. Prescription fill data provide patient-level exposure metrics that can further our understanding of pharmacogenomic medication utilization and help inform opportunities for pharmacogenomic testing.

On Inductive-Transductive Learning With Graph Neural Networks.

Many real-world domains involve information naturally represented by graphs, where nodes denote basic patterns while edges stand for relationships among them. The graph neural network (GNN) is a machine learning model capable of directly managing graph-structured data. In the original framework, GNNs are inductively trained, adapting their parameters based on a supervised learning environment. However, GNNs can also take advantage of transductive learning, thanks to the natural way they make information flow and spread across the graph, using relationships among patterns. In this paper, we propose a mixed inductive-transductive GNN model, study its properties and introduce an experimental strategy that allows us to understand and distinguish the role of inductive and transductive learning. The preliminary experimental results show interesting properties for the mixed model, highlighting how the peculiarities of the problems and the data can impact on the two learning strategies.

A Bioinformatics Approach to Investigate Structural and Non-Structural Proteins in Human Coronaviruses.

Recent studies confirmed that people unexposed to SARS-CoV-2 have preexisting reactivity, probably due to previous exposure to widely circulating common cold coronaviruses. Such preexistent reactivity against SARS-CoV-2 comes from memory T cells that can specifically recognize a SARS-CoV-2 epitope of structural and non-structural proteins and the homologous epitopes from common cold coronaviruses. Therefore, it is important to understand the SARS-CoV-2 cross-reactivity by investigating these protein sequence similarities with those of different circulating coronaviruses. In addition, the emerging SARS-CoV-2 variants lead to an intense interest in whether mutations in proteins (especially in the spike) could potentially compromise vaccine effectiveness. Since it is not clear that the differences in clinical outcomes are caused by common cold coronaviruses, a deeper investigation on cross-reactive T-cell immunity to SARS-CoV-2 is crucial to examine the differential COVID-19 symptoms and vaccine performance. Therefore, the present study can be a starting point for further research on cross-reactive T cell recognition between circulating common cold coronaviruses and SARS-CoV-2, including the most recent variants Delta and Omicron. In the end, a deep learning approach, based on Siamese networks, is proposed to accurately and efficiently calculate a BLAST-like similarity score between protein sequences.

Structural Bioinformatic Survey of Protein-Small Molecule Interfaces Delineates the Role of Glycine in Surface Pocket Formation.

With a structural bioinformatic approach, we have explored amino acid compositions at PISA defined interfaces between small molecules and proteins that are contained in an optimized subset of 11,351 PDB files. The use of a series of restrictions, to prevent redundancy and biases from interactions between amino acids with charged side chains and ions, yielded a final data set of 45,230 protein-small molecule interfaces. We have compared occurrences of natural amino acids in surface exposed regions and binding sites for all the proteins of our data set. From our structural bioinformatic survey, the most relevant signal arose from the unexpected Gly abundance at enzyme catalytic sites. This finding suggested that Gly must have a fundamental role in stabilizing concave protein surface moieties. Subsequently, we have tried to predict the effect of in silico Gly mutations in hen egg white lysozyme to optimize those conditions that can reshape the protein surface with the appearance of new pockets. Replacing amino acids having bulky side chains with Gly in specific protein regions seems a feasible way for designing proteins with additional surface pockets, which can alter protein surface dynamics, therefore, representing controllable switches for protein activity.

Structural bioinformatics survey on disease-inducing missense mutations.

Understanding the molecular mechanisms that correlate pathologies with missense mutations is of critical importance for disease risk estimations and for devising personalized therapies. Thus, we have performed a bioinformatic survey of ClinVar, a database of human genomic variations, to find signals that can account for missense mutation pathogenicity. Arginine resulted as the most frequently replaced amino acid both in benign and pathogenic mutations. By adding the structural dimension to this investigation to increase its resolution, we found that arginine mutations occurring at the protein-DNA interface increase pathogenicity 6.5 times with respect to benign variants. Glycine is the second amino acid among all the pathological missense mutations. Necessarily replaced by larger amino acids, glycine substitutions perturb the structural stability of proteins and, therefore, their functions, being mostly located in buried protein moieties. Arginine and glycine appear as representative of missense mutations causing respective changes in interaction processes and protein structural features, the two main molecular mechanisms of genome-induced pathologies.

Multi-Modal Siamese Network for Diagnostically Similar Lesion Retrieval in Prostate MRI.

Multi-parametric prostate MRI (mpMRI) is a powerful tool to diagnose prostate cancer, though difficult to interpret even for experienced radiologists. A common radiological procedure is to compare a magnetic resonance image with similarly diagnosed cases. To assist the radiological image interpretation process, computerized Content-Based Image Retrieval systems (CBIRs) can therefore be employed to improve the reporting workflow and increase its accuracy. In this article, we propose a new, supervised siamese deep learning architecture able to handle multi-modal and multi-view MR images with similar PIRADS score. An experimental comparison with well-established deep learning-based CBIRs (namely standard siamese networks and autoencoders) showed significantly improved performance with respect to both diagnostic (ROC-AUC), and information retrieval metrics (Precision-Recall, Discounted Cumulative Gain and Mean Average Precision). Finally, the new proposed multi-view siamese network is general in design, facilitating a broad use in diagnostic medical imaging retrieval.

A new deep learning approach integrated with clinical data for the dermoscopic differentiation of early melanomas from atypical nevi.

BACKGROUND: Timely recognition of malignant melanoma (MM) is challenging for dermatologists worldwide and represents the main determinant for mortality. Dermoscopic examination is influenced by dermatologists' experience and fails to achieve adequate accuracy and reproducibility in discriminating atypical nevi (AN) from early melanomas (EM). OBJECTIVE: We aimed to develop a Deep Convolutional Neural Network (DCNN) model able to support dermatologists in the classification and management of atypical melanocytic skin lesions (aMSL). METHODS: A training set (630 images), a validation set (135) and a testing set (214) were derived from the idScore dataset of 979 challenging aMSL cases in which the dermoscopic image is integrated with clinical data (age, sex, body site and diameter) and associated with histological data. A DCNN_aMSL architecture was designed and then trained on both dermoscopic images of aMSL and the clinical/anamnestic data, resulting in the integrated "iDCNN_aMSL" model. Responses of 111 dermatologists with different experience levels on both aMSL classification (intuitive diagnosis) and management decisions (no/long follow-up; short follow-up; excision/preventive excision) were compared with the DCNNs models. RESULTS: In the lesion classification study, the iDCNN_aMSL achieved the best accuracy, reaching an AUC = 90.3 %, SE = 86.5 % and SP = 73.6 %, compared to DCNN_aMSL (SE = 89.2 %, SP = 65.7 %) and intuitive diagnosis of dermatologists (SE = 77.0 %; SP = 61.4 %). CONCLUSIONS: The iDCNN_aMSL proved to be the best support tool for management decisions reducing the ratio of inappropriate excision. The proposed iDCNN_aMSL model can represent a valid support for dermatologists in discriminating AN from EM with high accuracy and for medical decision making by reducing their rates of inappropriate excisions.

Image generation by GAN and style transfer for agar plate image segmentation.

BACKGROUND AND OBJECTIVES: Deep learning models and specifically Convolutional Neural Networks (CNNs) are becoming the leading approach in many computer vision tasks, including medical image analysis. Nevertheless, the CNN training usually requires large sets of supervised data, which are often difficult and expensive to obtain in the medical field. To address the lack of annotated images, image generation is a promising method, which is becoming increasingly popular in the computer vision community. In this paper, we present a new approach to the semantic segmentation of bacterial colonies in agar plate images, based on deep learning and synthetic image generation, to increase the training set size. Indeed, semantic segmentation of bacterial colony is the basis for infection recognition and bacterial counting in Petri plate analysis. METHODS: A convolutional neural network (CNN) is used to separate the bacterial colonies from the background. To face the lack of annotated images, a novel engine is designed - which exploits a generative adversarial network to capture the typical distribution of the bacterial colonies on agar plates - to generate synthetic data. Then, bacterial colony patches are superimposed on existing background images, taking into account both the local appearance of the background and the intrinsic opacity of the bacterial colonies, and a style transfer algorithm is used for further improve visual realism. RESULTS: The proposed deep learning approach has been tested on the only public dataset available with pixel-level annotations for bacterial colony semantic segmentation in agar plates. The role of including synthetic data in the training of a segmentation CNN has been evaluated, showing how comparable performances can be obtained with respect to the use of real images. Qualitative results are also reported for a second public dataset in which the segmentation annotations are not provided. CONCLUSIONS: The use of a small set of real data, together with synthetic images, allows obtaining comparable results with respect to using a complete set of real images. Therefore, the proposed synthetic data generator is able to address the scarcity of biomedical data and provides a scalable and cheap alternative to human ground-truth supervision.