Machine learning-based prediction of adverse drug effects: An example of seizure-inducing compounds

A primer on the use of machine learning to distil knowledge from data in biological psychiatry

Applications of machine learning in the biomedical sciences are growing rapidly. This growth has been spurred by diverse cross-institutional and interdisciplinary collaborations, public availability of large datasets, an increase in the accessibility of analytic routines, and the availability of powerful computing resources. With this increased access and exposure to machine learning comes a responsibility for education and a deeper understanding of its bases and bounds, borne equally by data scientists seeking to ply their analytic wares in medical research and by biomedical scientists seeking to harness such methods to glean knowledge from data. This article provides an accessible and critical review of machine learning for a biomedically informed audience, as well as its applications in psychiatry. The review covers definitions and expositions of commonly used machine learning methods, and historical trends of their use in psychiatry. We also provide a set of standards, namely Guidelines for REporting Machine Learning Investigations in Neuropsychiatry (GREMLIN), for designing and reporting studies that use machine learning as a primary data-analysis approach. Lastly, we propose the establishment of the Machine Learning in Psychiatry (MLPsych) Consortium, enumerate its objectives, and identify areas of opportunity for future applications of machine learning in biological psychiatry. This review serves as a cautiously optimistic primer on machine learning for those on the precipice as they prepare to dive into the field, either as methodological practitioners or well-informed consumers.

Comparative assessment of Ca2+ oscillations in 2- and 3-dimensional hiPSC derived and isolated cortical neuronal networks

Human induced Pluripotent Stem Cell (hiPSC) derived neural cells offer great potential for modelling neurological diseases and toxicities and have found application in drug discovery and toxicology. As part of the European Innovative Medicines Initiative (IMI2) NeuroDeRisk (Neurotoxicity De-Risking in Preclinical Drug Discovery), we here explore the Ca2+ oscillation responses of 2D and 3D hiPSC derived neuronal networks of mixed Glutamatergic/GABAergic activity with a compound set encompassing both clinically as well as experimentally determined seizurogenic compounds. Both types of networks are scored against Ca2+ responses of a primary mouse cortical neuronal 2D network model serving as an established comparator assay. Parameters of frequency and amplitude of spontaneous global network Ca2+ oscillations and the drug-dependent directional changes to these were assessed, and predictivity of seizurogenicity scored using contingency table analysis. In addition, responses between models were compared between both 2D models as well as between 2D and 3D models. Concordance of parameter responses was best between the hiPSC neurospheroid and the mouse primary cortical neuron model (77% for frequency and 65% for amplitude). Decreases in spontaneous Ca2+ oscillation frequency and amplitude were found to be the most basic shared determinants of risk of seizurogenicity between the mouse and the neurospheroid model based on testing of clinical compounds with documented seizurogenic activity. Increases in spontaneous Ca2+ oscillation frequency were primarily observed with the 2D hIPSC model, though the specificity of this effect to seizurogenic clinical compounds was low (33%), while decreases to spike amplitude in this model were more predictive of seizurogenicity. Overall predictivities of the models were similar, with sensitivity of the assays typically exceeding specificity due to high false positive rates. Higher concordance of the hiPSC 3D model over the 2D model when compared to mouse cortical 2D responses may be the result of both a longer maturation time of the neurospheroid (84-87 days for 3D vs. 22-24 days for 2D maturation) as well as the 3-dimensional nature of network connections established. The simplicity and reproducibility of spontaneous Ca2+ oscillation readouts support further investigation of hiPSC derived neuronal sources and their 2- and 3-dimensional networks for neuropharmacological safety screening.

Comparative study for the IMI2-NeuroDeRisk project on microelectrode arrays to derisk drug-induced seizure liability

INTRODUCTION

In the framework of the IMI2-NeuroDeRisk consortium, three in vitro electrophysiology assays were compared to improve preclinical prediction of seizure-inducing liabilities.

METHODS

Two cell models, primary rat cortical neurons and human induced pluripotent stem cell (hiPSC)-derived glutamatergic neurons co-cultured with hiPSC-derived astrocytes were tested on two different microelectrode array (MEA) platforms, Maestro Pro (Axion Biosystems) and Multiwell-MEA-System (Multi Channel Systems), in three separate laboratories. Pentylenetetrazole (PTZ) and/or picrotoxin (PTX) were included in each plate as positive (n = 3-6 wells) and ≤0.2% DMSO was used as negative controls (n = 3-12 wells). In general, concentrations in a range of 0.1-30 μM were tested, anchored, when possible, on clinically relevant exposures (unbound Cmax) were tested. Activity thresholds for drug-induced changes were set at 20%. To evaluate sensitivity, specificity and predictivity of the cell models, seizurogenic responses were defined as changes in 4 or more endpoints. Concentration dependence trends were also considered.

RESULTS

Neuronal activity of 33 compounds categorized as positive tool drugs, seizure-positive or seizure-negative compounds was evaluated. Acute drug effects (<60 min) were compared to baseline recordings. Time points < 15 min exhibited stronger, less variable responses to many of the test agents. For many compounds a reduction and cessation of neuronal activity was detected at higher test concentrations. There was not a single pattern of seizurogenic activity detected, even among tool compounds, likely due to different mechanisms of actions and/or off-target profiles. A post-hoc analysis focusing on changes indicative of neuronal excitation is presented.

CONCLUSION

All cell models showed good sensitivity, ranging from 70 to 86%. Specificity ranged from 40 to 70%. Compared to more conventional measurements of evoked activity in hippocampal slices, these plate-based models provide higher throughput and the potential to study subacute responses. Yet, they may be limited by the random, spontaneous nature of their network activity.

Assessing seizure liability in vitro with voltage-sensitive dye imaging in mouse hippocampal slices

Non-clinical toxicology is a major cause of drug candidate attrition during development. In particular, drug-induced seizures are the most common finding in central nervous system (CNS) toxicity. Current safety pharmacology tests for assessing CNS functions are often inadequate in detecting seizure-inducing compounds early in drug development, leading to significant delays. This paper presents an in vitro seizure liability assay using voltage-sensitive dye (VSD) imaging techniques in hippocampal brain slices, offering a powerful alternative to traditional electrophysiological methods. Hippocampal slices were isolated from mice, and VSD optical responses evoked by stimulating the Schaffer collateral pathway were recorded and analyzed in the stratum radiatum (SR) and stratum pyramidale (SP). VSDs allow for the comprehensive visualization of neuronal action potentials and postsynaptic potentials on a millisecond timescale. By employing this approach, we investigated the in vitro drug-induced seizure liability of representative pro-convulsant compounds. Picrotoxin (PiTX; 1-100 μM), gabazine (GZ; 0.1-10 μM), and 4-aminopyridine (4AP; 10-100 μM) exhibited seizure-like responses in the hippocampus, but pilocarpine hydrochloride (Pilo; 10-100 μM) did not. Our findings demonstrate the potential of VSD-based assays in identifying seizurogenic compounds during early drug discovery, thereby reducing delays in drug development and providing insights into the mechanisms underlying seizure induction and the associated risks of pro-convulsant compounds.

The value of machine learning for prognosis prediction of diphenhydramine exposure: National analysis of 50,000 patients in the United States

Background

Diphenhydramine (DPH) is an antihistamine medication that in overdose can result in anticholinergic symptoms and serious complications, including arrhythmia and coma. We aimed to compare the value of various machine learning (ML) models, including light gradient boosting machine (LGBM), logistic regression (LR), and random forest (RF), in the outcome prediction of DPH poisoning.

Materials and Methods

We used the National Poison Data System database and included all of the human exposures of DPH from January 01, 2017 to December 31, 2017, and excluded those cases with missing information, duplicated cases, and those who reported co-ingestion. Data were split into training and test datasets, and three ML models were compared. We developed confusion matrices for each, and standard performance metrics were calculated.

Results

Our study population included 53,761 patients with DPH exposure. The most common reasons for exposure, outcome, chronicity of exposure, and formulation were captured. Our results showed that the average precision-recall area under the curve (AUC) of 0.84. LGBM and RF had the highest performance (average AUC of 0.91), followed by LR (average AUC of 0.90). The specificity of the models was 87.0% in the testing groups. The precision of models was 75.0%. Recall (sensitivity) of models ranged between 73% and 75% with an F1 score of 75.0%. The overall accuracy of LGBM, LR, and RF models in the test dataset was 74.8%, 74.0%, and 75.1%, respectively. In total, just 1.1% of patients (mostly those with major outcomes) received physostigmine.

Conclusion

Our study demonstrates the application of ML in the prediction of DPH poisoning.

Gene Expression Profile as a Predictor of Seizure Liability

Analysis platforms to predict drug-induced seizure liability at an early phase of drug development would improve safety and reduce attrition and the high cost of drug development. We hypothesized that a drug-induced in vitro transcriptomics signature predicts its ictogenicity. We exposed rat cortical neuronal cultures to non-toxic concentrations of 34 compounds for 24 h; 11 were known to be ictogenic (tool compounds), 13 were associated with a high number of seizure-related adverse event reports in the clinical FDA Adverse Event Reporting System (FAERS) database and systematic literature search (FAERS-positive compounds), and 10 were known to be non-ictogenic (FAERS-negative compounds). The drug-induced gene expression profile was assessed from RNA-sequencing data. Transcriptomics profiles induced by the tool, FAERS-positive and FAERS-negative compounds, were compared using bioinformatics and machine learning. Of the 13 FAERS-positive compounds, 11 induced significant differential gene expression; 10 of the 11 showed an overall high similarity to the profile of at least one tool compound, correctly predicting the ictogenicity. Alikeness-% based on the number of the same differentially expressed genes correctly categorized 85%, the Gene Set Enrichment Analysis score correctly categorized 73%, and the machine-learning approach correctly categorized 91% of the FAERS-positive compounds with reported seizure liability currently in clinical use. Our data suggest that the drug-induced gene expression profile could be used as a predictive biomarker for seizure liability.

Early Differentiation of Irreversible Electroporation Ablation Regions With Radiomics Features of Conventional MRI

RATIONALE AND OBJECTIVES

Irreversible electroporation (IRE) is a promising non-thermal ablation technique for the treatment of patients with hepatocellular carcinoma. Early differentiation of the IRE zone from surrounding reversibly electroporated (RE) penumbra is vital for the evaluation of treatment response. In this study, an advanced statistical learning framework was developed by evaluating standard MRI data to differentiate IRE ablation zones, and to correlate with histological tumor biomarkers.

MATERIALS AND METHODS

Fourteen rabbits with VX2 liver tumors were scanned following IRE ablation and forty-six features were extracted from T1w and T2w MRI. Following identification of key imaging variables through two-step feature analysis, multivariable classification and regression models were generated for differentiation of IRE ablation zones, and correlation with histological markers reflecting viable tumor cells, microvessel density, and apoptosis rate. The performance of the multivariable models was assessed by measuring accuracy, receiver operating characteristics curve analysis, and Spearman correlation coefficients.

RESULTS

The classifiers integrating four radiomics features of T1w, T2w, and T1w+T2w MRI data distinguished IRE from RE zones with an accuracy of 97%, 80%, and 97%, respectively. Also, pixelwise classification models of T1w, T2w, and T1w+T2w MRI labeled each voxel with an accuracy of 82.8%, 66.5%, and 82.9%, respectively. Regression models obtained a strong correlation with behavior of viable tumor cells (0.62 ≤ r2 ≤ 0.85, p < 0.01), apoptosis (0.40 ≤ r2 ≤ 0.82, p < 0.01), and microvessel density (0.48 ≤ r2 ≤ 0.58, p < 0.01).

CONCLUSION

MRI radiomics features provide descriptive power for early differentiation of IRE and RE zones while observing strong correlations among multivariable MRI regression models and histological tumor biomarkers.

Blockchain technology: A DNN token-based approach in healthcare and COVID-19 to generate extracted data

The healthcare technologies in COVID-19 pandemic had grown immensely in various domains. Blockchain technology is one such turnkey technology, which is transforming the data securely; to store electronic health records (EHRs), develop deep learning algorithms, access the data, process the data between physicians and patients to access the EHRs in the form of distributed ledgers. Blockchain technology is also made to supply the data in the cloud and contact the huge amount of healthcare data, which is difficult and complex to process. As the complexity in the analysis of data is increasing day by day, it has become essential to minimize the risk of data complexity. This paper supports deep neural network (DNN) analysis in healthcare and COVID-19 pandemic and gives the smart contract procedure, to identify the feature extracted data (FED) from the existing data. At the same time, the innovation will be useful to analyse future diseases. The proposed method also analyze the existing diseases which had been reported and it is extremely useful to guide physicians in providing appropriate treatment and save lives. To achieve this, the massive data is integrated using Python scripting language under various libraries to perform a wide range of medical and healthcare functions to infer knowledge that assists in the diagnosis of major diseases such as heart disease, blood cancer, gastric and COVID-19.

Raster plots machine learning to predict the seizure liability of drugs and to identify drugs

In vitro microelectrode array (MEA) assessment using human induced pluripotent stem cell (iPSC)-derived neurons holds promise as a method of seizure and toxicity evaluation. However, there are still issues surrounding the analysis methods used to predict seizure and toxicity liability as well as drug mechanisms of action. In the present study, we developed an artificial intelligence (AI) capable of predicting the seizure liability of drugs and identifying drugs using deep learning based on raster plots of neural network activity. The seizure liability prediction AI had a prediction accuracy of 98.4% for the drugs used to train it, classifying them correctly based on their responses as either seizure-causing compounds or seizure-free compounds. The AI also made concentration-dependent judgments of the seizure liability of drugs that it was not trained on. In addition, the drug identification AI implemented using the leave-one-sample-out scheme could distinguish among 13 seizure-causing compounds as well as seizure-free compound responses, with a mean accuracy of 99.9 ± 0.1% for all drugs. These AI prediction models are able to identify seizure liability concentration-dependence, rank the level of seizure liability based on the seizure liability probability, and identify the mechanism of the action of compounds. This holds promise for the future of in vitro MEA assessment as a powerful, high-accuracy new seizure liability prediction method.

Sensitivity, specificity and limitation of in vitro hippocampal slice and neuron-based assays for assessment of drug-induced seizure liability

An effective in vitro screening assay to detect seizure liability in preclinical development can contribute to better lead molecule optimization prior to candidate selection, providing higher throughput and overcoming potential brain exposure limitations in animal studies. This study explored effects of 26 positive and 14 negative reference pharmacological agents acting through different mechanisms, including 18 reference agents acting on glutamate signaling pathways, in a brain slice assay (BSA) of adult rat to define the assay's sensitivity, specificity, and limitations. Evoked population spikes (PS) were recorded from CA1 pyramidal neurons of hippocampus (HPC) in the BSA. Endpoints for analysis were PS area and PS number. Most positive references (24/26) elicited a concentration-dependent increase in PS area and/or PS number. The negative references (14/14) had little effect on the PS. Moreover, we studied the effects of 15 reference agents testing positive in the BSA on spontaneous activity in E18 rat HPC neurons monitored with microelectrode arrays (MEA), and compared these effects to the BSA results. From these in vitro studies we conclude that the BSA provides 93% sensitivity and 100% specificity in prediction of drug-induced seizure liability, including detecting seizurogenicity by 3 groups of metabotropic glutamate receptor (mGluR) ligands. The MEA results seemed more variable, both quantitatively and directionally, particularly for endpoints capturing synchronized electrical activity. We discuss these results from the two models, comparing each with published results, and provide potential explanations for differences and future directions.

Accelerating antibiotic discovery through artificial intelligence

By targeting invasive organisms, antibiotics insert themselves into the ancient struggle of the host-pathogen evolutionary arms race. As pathogens evolve tactics for evading antibiotics, therapies decline in efficacy and must be replaced, distinguishing antibiotics from most other forms of drug development. Together with a slow and expensive antibiotic development pipeline, the proliferation of drug-resistant pathogens drives urgent interest in computational methods that promise to expedite candidate discovery. Strides in artificial intelligence (AI) have encouraged its application to multiple dimensions of computer-aided drug design, with increasing application to antibiotic discovery. This review describes AI-facilitated advances in the discovery of both small molecule antibiotics and antimicrobial peptides. Beyond the essential prediction of antimicrobial activity, emphasis is also given to antimicrobial compound representation, determination of drug-likeness traits, antimicrobial resistance, and de novo molecular design. Given the urgency of the antimicrobial resistance crisis, we analyze uptake of open science best practices in AI-driven antibiotic discovery and argue for openness and reproducibility as a means of accelerating preclinical research. Finally, trends in the literature and areas for future inquiry are discussed, as artificially intelligent enhancements to drug discovery at large offer many opportunities for future applications in antibiotic development.

Spectrum of deep learning algorithms in drug discovery

Deep learning (DL) algorithms are a subset of machine learning algorithms with the aim of modeling complex mapping between a set of elements and their classes. In parallel to the advance in revealing the molecular bases of diseases, a notable innovation has been undertaken to apply DL in data/libraries management, reaction optimizations, differentiating uncertainties, molecule constructions, creating metrics from qualitative results, and prediction of structures or interactions. From source identification to lead discovery and medicinal chemistry of the drug candidate, drug delivery, and modification, the challenges can be subjected to artificial intelligence algorithms to aid in the generation and interpretation of data. Discovery and design approach, both demand automation, large data management and data fusion by the advance in high-throughput mode. The application of DL can accelerate the exploration of drug mechanisms, finding novel indications for existing drugs (drug repositioning), drug development, and preclinical and clinical studies. The impact of DL in the workflow of drug discovery, design, and their complementary tools are highlighted in this review. Additionally, the type of DL algorithms used for this purpose, and their pros and cons along with the dominant directions of future research are presented.

Early Drug Development and Evaluation of Putative Antitubercular Compounds in the -Omics Era

Tuberculosis (TB) is an infectious disease caused by the bacterium Mycobacterium tuberculosis. According to the WHO, the disease is one of the top 10 causes of death of people worldwide. Mycobacterium tuberculosis is an intracellular pathogen with an unusually thick, waxy cell wall and a complex life cycle. These factors, combined with M. tuberculosis ability to enter prolonged periods of latency, make the bacterium very difficult to eradicate. The standard treatment of TB requires 6-20months, depending on the drug susceptibility of the infecting strain. The need to take cocktails of antibiotics to treat tuberculosis effectively and the emergence of drug-resistant strains prompts the need to search for new antitubercular compounds. This review provides a perspective on how modern -omic technologies facilitate the drug discovery process for tuberculosis treatment. We discuss how methods of DNA and RNA sequencing, proteomics, and genetic manipulation of organisms increase our understanding of mechanisms of action of antibiotics and allow the evaluation of drugs. We explore the utility of mathematical modeling and modern computational analysis for the drug discovery process. Finally, we summarize how -omic technologies contribute to our understanding of the emergence of drug resistance.

Machine Learning in Predictive Toxicology: Recent Applications and Future Directions for Classification Models

In recent times, machine learning has become increasingly prominent in predictive toxicology as it has shifted from in vivo studies toward in silico studies. Currently, in vitro methods together with other computational methods such as quantitative structure-activity relationship modeling and absorption, distribution, metabolism, and excretion calculations are being used. An overview of machine learning and its applications in predictive toxicology is presented here, including support vector machines (SVMs), random forest (RF) and decision trees (DTs), neural networks, regression models, naïve Bayes, k-nearest neighbors, and ensemble learning. The recent successes of these machine learning methods in predictive toxicology are summarized, and a comparison of some models used in predictive toxicology is presented. In predictive toxicology, SVMs, RF, and DTs are the dominant machine learning methods due to the characteristics of the data available. Lastly, this review describes the current challenges facing the use of machine learning in predictive toxicology and offers insights into the possible areas of improvement in the field.

Where are all the anthelmintics? Challenges and opportunities on the path to new anthelmintics

Control of helminth parasites is a key challenge for human and veterinary medicine. In the absence of effective vaccines and adequate sanitation, prophylaxis and treatment commonly rely upon anthelmintics. There are concerns about the development of drug resistance, side-effects, lack of efficacy and cost-effectiveness that drive the need for new classes of anthelmintics. Despite this need, only three new drug classes have reached the animal market since 2000 and no new classes of anthelmintic have been approved for human use. So where are all the anthelmintics? What are the barriers to anthelmintic discovery, and what emerging opportunities can be used to address this? This was a discussion group focus at the 2019 8th Consortium for Anthelmintic Resistance and Susceptibility (CARS) in Wisconsin, USA. Here we report the findings of the group in the broader context of the human and veterinary anthelmintic discovery pipeline, highlighting challenges unique to antiparasitic drug discovery. We comment on why the development of novel anthelmintics has been so rare. Further, we discuss potential opportunities for drug development moving into the 21st Century.

Applicability of hiPSC-Derived Neuronal Cocultures and Rodent Primary Cortical Cultures for In Vitro Seizure Liability Assessment

Seizures are life-threatening adverse drug reactions which are investigated late in drug development using rodent models. Consequently, if seizures are detected, a lot of time, money and animals have been used. Thus, there is a need for in vitro screening models using human cells to circumvent interspecies translation. We assessed the suitability of cocultures of human-induced pluripotent stem cell (hiPSC)-derived neurons and astrocytes compared with rodent primary cortical cultures for in vitro seizure liability assessment using microelectrode arrays. hiPSC-derived and rodent primary cortical neuronal cocultures were exposed to 9 known (non)seizurogenic compounds (pentylenetetrazole, amoxapine, enoxacin, amoxicillin, linopirdine, pilocarpine, chlorpromazine, phenytoin, and acetaminophen) to assess effects on neuronal network activity using microelectrode array recordings. All compounds affect activity in hiPSC-derived cocultures. In rodent primary cultures all compounds, except amoxicillin changed activity. Changes in activity patterns for both cell models differ for different classes of compounds. Both models had a comparable sensitivity for exposure to amoxapine (lowest observed effect concentration [LOEC] 0.03 µM), linopirdine (LOEC 1 µM), and pilocarpine (LOEC 0.3 µM). However, hiPSC-derived cultures were about 3 times more sensitive for exposure to pentylenetetrazole (LOEC 30 µM) than rodent primary cortical cultures (LOEC 100 µM). Sensitivity of hiPSC-derived cultures for chlorpromazine, phenytoin, and enoxacin was 10-30 times higher (LOECs 0.1, 0.3, and 0.1 µM, respectively) than in rodent cultures (LOECs 10, 3, and 3 µM, respectively). Our data indicate that hiPSC-derived neuronal cocultures may outperform rodent primary cortical cultures with respect to detecting seizures, thereby paving the way towards animal-free seizure assessment.

[Strong demands for the new preclinical assessment system of the CNS adverse effects]

Human induced pluripotent stem cell-derived neurons (hiPSC-neurons) have potentials to improve the predictability of the adverse effects (AEs) in the human central nervous system (CNS). We have succeeded in reproducing the human neural networks stably on dish. This is important because most of the CNS AEs were caused not by the neuronal death but by the abnormal neural network activities. Our group have been attempting to establish the in vitro assay system to detect abnormal neural network activities by microelectrode array system (MEA). Furthermore, the algorithm that can determine the synchronized burst firing (SBF) objectively was developed in the collaboration of our iPS Non-clinical Experiments for Nervous System (iNCENS) project and Consortium for Safety Assessment using Human iPS Cells (CSAHi). iNCENS and CSAHi are also involved in the international Life Sciences Institute (ILSI): Health and Environmental Sciences Institute (HESI) HESI NeuTox MEA sub team, and are participating in the pilot study using 12 compounds. In the big flow to develop physiological CNS safety assessment system, we have to show Japan initiative based on our innovative iPS technology we have developed so far.

Digital microbiology

BACKGROUND

Digitalization and artificial intelligence have an important impact on the way microbiology laboratories will work in the near future. Opportunities and challenges lie ahead to digitalize the microbiological workflows. Making efficient use of big data, machine learning, and artificial intelligence in clinical microbiology requires a profound understanding of data handling aspects.

OBJECTIVE

This review article summarizes the most important concepts of digital microbiology. The article gives microbiologists, clinicians and data scientists a viewpoint and practical examples along the diagnostic process.

SOURCES

We used peer-reviewed literature identified by a PubMed search for digitalization, machine learning, artificial intelligence and microbiology.

CONTENT

We describe the opportunities and challenges of digitalization in microbiological diagnostic processes with various examples. We also provide in this context key aspects of data structure and interoperability, as well as legal aspects. Finally, we outline the way for applications in a modern microbiology laboratory.

IMPLICATIONS

We predict that digitalization and the usage of machine learning will have a profound impact on the daily routine of laboratory staff. Along the analytical process, the most important steps should be identified, where digital technologies can be applied and provide a benefit. The education of all staff involved should be adapted to prepare for the advances in digital microbiology.

Comparison of the predictive outcomes for anti-tuberculosis drug-induced hepatotoxicity by different machine learning techniques

BACKGROUND

The study compared the predictive outcomes of artificial neural network, support vector machine and random forest on the occurrence of anti-tuberculosis drug-induced hepatotoxicity.

METHODS

The clinical and genomic data of patients treated with anti-tuberculosis drugs at Taipei Medical University-Wanfang Hospital were used as training sets, and those at Taipei Medical University-Shuang Ho Hospital served as test sets. Features were selected through a univariate risk factor analysis and literature evaluation. The accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve were calculated to compare the traditional, genomic, and combined models of the three techniques.

RESULTS

Nine models were created with 7 clinical factors and 4 genotypes. Artificial neural network with clinical and genomic factors exhibited the best performance, with an accuracy of 88.67%, a sensitivity of 80%, and a specificity of 90.4% for the test set. The area under the receiver operating characteristic curve of this best model reached 0.894 for training set and 0.898 for test set, which was significantly better than 0.801 for training set and 0.728 for test set by support vector machine and 0.724 for training set and 0.718 for test set by random forest.

CONCLUSIONS

Artificial neural network with clinical and genomic data can become a clinical useful tool in predicting anti-tuberculosis drug-induced hepatotoxicity. The machine learning technique can be an innovation to predict and prevent adverse drug reaction.

A drug identification model developed using deep learning technologies: experience of a medical center in Taiwan

BACKGROUND

Issuing of correct prescriptions is a foundation of patient safety. Medication errors represent one of the most important problems in health care, with 'look-alike and sound-alike' (LASA) being the lead error. Existing solutions to prevent LASA still have their limitations. Deep learning techniques have revolutionized identification classifiers in many fields. In search of better image-based solutions for blister package identification problem, this study using a baseline deep learning drug identification (DLDI) aims to understand how identification confusion of look-alike images by human occurs through the cognitive counterpart of deep learning solutions and thereof to suggest further solutions to approach them.

METHODS

We collected images of 250 types of blister-packaged drug from the Out-Patient Department (OPD) of a medical center for identification. The deep learning framework of You Only Look Once (YOLO) was adopted for implementation of the proposed deep learning. The commonly-used F1 score, defined by precision and recall for large numbers of identification tests, was used as the performance criterion. This study trained and compared the proposed models based on images of either the front-side or back-side of blister-packaged drugs.

RESULTS

Our results showed that the total training time for the front-side model and back-side model was 5 h 34 min and 7 h 42 min, respectively. The F1 score of the back-side model (95.99%) was better than that of the front-side model (93.72%).

CONCLUSIONS

In conclusion, this study constructed a deep learning-based model for blister-packaged drug identification, with an accuracy greater than 90%. This model outperformed identification using conventional computer vision solutions, and could assist pharmacists in identifying drugs while preventing medication errors caused by look-alike blister packages. By integration into existing prescription systems in hospitals, the results of this study indicated that using this model, drugs dispensed could be verified in order to achieve automated prescription and dispensing.

In silico toxicology: From structure–activity relationships towards deep learning and adverse outcome pathways

In silico toxicology is an emerging field. It gains increasing importance as research is aiming to decrease the use of animal experiments as suggested in the 3R principles by Russell and Burch. In silico toxicology is a means to identify hazards of compounds before synthesis, and thus in very early stages of drug development. For chemical industries, as well as regulatory agencies it can aid in gap‐filling and guide risk minimization strategies. Techniques such as structural alerts, read‐across, quantitative structure–activity relationship, machine learning, and deep learning allow to use in silico toxicology in many cases, some even when data is scarce. Especially the concept of adverse outcome pathways puts all techniques into a broader context and can elucidate predictions by mechanistic insights.

This article is categorized under:

Structure and Mechanism > Computational Biochemistry and Biophysics

Data Science > Chemoinformatics

Deep Learning-Based Structure-Activity Relationship Modeling for Multi-Category Toxicity Classification: A Case Study of 10K Tox21 Chemicals With High-Throughput Cell-Based Androgen Receptor Bioassay Data

Deep learning (DL) has attracted the attention of computational toxicologists as it offers a potentially greater power for in silico predictive toxicology than existing shallow learning algorithms. However, contradicting reports have been documented. To further explore the advantages of DL over shallow learning, we conducted this case study using two cell-based androgen receptor (AR) activity datasets with 10K chemicals generated from the Tox21 program. A nested double-loop cross-validation approach was adopted along with a stratified sampling strategy for partitioning chemicals of multiple AR activity classes (i.e., agonist, antagonist, inactive, and inconclusive) at the same distribution rates amongst the training, validation and test subsets. Deep neural networks (DNN) and random forest (RF), representing deep and shallow learning algorithms, respectively, were chosen to carry out structure-activity relationship-based chemical toxicity prediction. Results suggest that DNN significantly outperformed RF (p < 0.001, ANOVA) by 22–27% for four metrics (precision, recall, F-measure, and AUPRC) and by 11% for another (AUROC). Further in-depth analyses of chemical scaffolding shed insights on structural alerts for AR agonists/antagonists and inactive/inconclusive compounds, which may aid in future drug discovery and improvement of toxicity prediction modeling.

Prediction of irinotecan toxicity in metastatic colorectal cancer patients based on machine learning models with pharmacokinetic parameters

Irinotecan (CPT-11) is a drug used against a wide variety of tumors, which can cause severe toxicity, possibly leading to the delay or suspension of the cycle, with the consequent impact on the prognosis of survival. The main goal of this work is to predict the toxicities derived from CPT-11 using artificial intelligence methods. The data for this study is conformed of 53 cycles of FOLFIRINOX, corresponding to patients with metastatic colorectal cancer. Supported by several demographic data, blood markers and pharmacokinetic parameters resulting from a non-compartmental pharmacokinetic study of CPT-11 and its metabolites (SN-38 and SN-38-G), we use machine learning techniques to predict high degrees of different toxicities (leukopenia, neutropenia and diarrhea) in new patients. We predict high degree of leukopenia with an accuracy of 76%, neutropenia with 75% and diarrhea with 91%. Among other variables, this study shows that the areas under the curve of CPT-11, SN-38 and SN-38-G play a relevant role in the prediction of the studied toxicities. The presented models allow to predict the degree of toxicity for each cycle of treatment according to the particularities of each patient.

Deep learning in pharmacogenomics: from gene regulation to patient stratification

This Perspective provides examples of current and future applications of deep learning in pharmacogenomics, including: identification of novel regulatory variants located in noncoding domains of the genome and their function as applied to pharmacoepigenomics; patient stratification from medical records; and the mechanistic prediction of drug response, targets and their interactions. Deep learning encapsulates a family of machine learning algorithms that has transformed many important subfields of artificial intelligence over the last decade, and has demonstrated breakthrough performance improvements on a wide range of tasks in biomedicine. We anticipate that in the future, deep learning will be widely used to predict personalized drug response and optimize medication selection and dosing, using knowledge extracted from large and complex molecular, epidemiological, clinical and demographic datasets.

Can a collaborative healthcare network improve the care of people with epilepsy?

New opportunities are now available to improve care in ways not possible previously. Information contained in electronic medical records can now be shared without identifying patients. With network collaboration, large numbers of medical records can be searched to identify patients most like the one whose complex medical situation challenges the physician. The clinical effectiveness of different treatment strategies can be assessed rapidly to help the clinician decide on the best treatment for this patient. Other capabilities from different components of the network can prompt the recognition of what is the best available option and encourage the sharing of information about programs and electronic tools. Difficulties related to privacy, harmonization, integration, and costs are expected, but these are currently being addressed successfully by groups of organizations led by those who recognize the benefits.

Converging blockchain and next-generation artificial intelligence technologies to decentralize and accelerate biomedical research and healthcare

The increased availability of data and recent advancements in artificial intelligence present the unprecedented opportunities in healthcare and major challenges for the patients, developers, providers and regulators. The novel deep learning and transfer learning techniques are turning any data about the person into medical data transforming simple facial pictures and videos into powerful sources of data for predictive analytics. Presently, the patients do not have control over the access privileges to their medical records and remain unaware of the true value of the data they have. In this paper, we provide an overview of the next-generation artificial intelligence and blockchain technologies and present innovative solutions that may be used to accelerate the biomedical research and enable patients with new tools to control and profit from their personal data as well with the incentives to undergo constant health monitoring. We introduce new concepts to appraise and evaluate personal records, including the combination-, time- and relationship-value of the data. We also present a roadmap for a blockchain-enabled decentralized personal health data ecosystem to enable novel approaches for drug discovery, biomarker development, and preventative healthcare. A secure and transparent distributed personal data marketplace utilizing blockchain and deep learning technologies may be able to resolve the challenges faced by the regulators and return the control over personal data including medical records back to the individuals.