# Development of a web-based platform implementing novel Predictor of Toxicity for Medical Devices (PredTox/MD) > **NIH NIH R44** · PREDICTIVE, LLC · 2024 · $898,248 ## Abstract 1 Medical devices contain chemicals that can leach and cause adverse effects. International standards (ISO 2 10993) require the evaluation of such chemicals for specific toxicity endpoints, including skin sensitization, 3 irritation, and cytotoxicity. Short-terms assays commonly used for this task are time-consuming, expensive, and 4 require the sacrifice of many animals. Emerging FDA directives call to restrict and, eventually, eliminate animal 5 testing of medical and cosmetic products and develop alternative methods including computational tools. To 6 address this unmet need, in Phase I of this project we have created the largest carefully curated and publicly 7 available Guinea Pig Maximization Test (GPMT) dataset and developed first-in-class machine learning models 8 that predict the GPMT outcome. We implemented our models within the fully operational Predictor of Skin 9 Sensitization for Medical Devices (PreSS/MD) web portal. In Phase II, we will create new models and software 10 modules for reliable assessment of chemicals found in medical devices for sensitization, irritation, and 11 cytotoxicity per ISO 10993 guidance. These modules will be both available for licensing as standalone tools or 12 web applications as well as integrated into novel Predictor of Toxicity for Medical Devices (PredTox/MD) web 13 portal. The proposed R&D studies are structured around the following Specific Aims: Specific Aim 1: Develop 14 a highly curated, comprehensive PredTox/MD database. We will collect, thoroughly curate, and integrate 15 public data for all human, in vivo, and in vitro regulatory assays for skin sensitization, irritation/corrosion, and 16 cytotoxicity. We will extend our database to include all available data on chemical mixtures and develop special 17 curation workflows to handle mixtures of any composition. Specific Aim 2: Develop validated computational 18 models to predict sensitization, irritation, and cytotoxicity for chemicals leaching from medical devices. 19 We will employ our widely accepted predictive Quantitative Structure-Activity Relationship (QSAR) modeling 20 workflow fully compliant with OECD model validation principles. Consensus ensemble models will be developed 21 with several descriptor types and machine learning algorithms, including deep and active learning and a 22 Bayesian model integrating multiple individual assay models to predict the overall chemical safety. Specific Aim 23 3: Develop software modules for assessing medical device toxicity and incorporate these modules into 24 PredTox/MD portal. Models and workflows developed in Aim 2 will be programmed as autonomous software 25 modules that will be integrated into PredTox/MD platform and available for individual licensing to enable rapid 26 multi-point toxicity assessment for extractables and leachables found in medical devices. Successful 27 completion of Phase II studies will result in the novel computational toolkit and web-based resource to 28 evaluat... ## Key facts - **NIH application ID:** 10820280 - **Project number:** 2R44ES032371-02A1 - **Recipient organization:** PREDICTIVE, LLC - **Principal Investigator:** Alexander Golbraikh - **Activity code:** R44 (R01, R21, SBIR, etc.) - **Funding institute:** NIH - **Fiscal year:** 2024 - **Award amount:** $898,248 - **Award type:** 2 - **Project period:** 2020-09-09 → 2025-12-31 ## Primary source NIH RePORTER: https://reporter.nih.gov/project-details/10820280 ## Citation > US National Institutes of Health, RePORTER application 10820280, Development of a web-based platform implementing novel Predictor of Toxicity for Medical Devices (PredTox/MD) (2R44ES032371-02A1). Retrieved via AI Analytics 2026-05-25 from https://api.ai-analytics.org/grant/nih/10820280. Licensed CC0. --- *[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*