CardioSim
Tox CML
AI-powered in silico chemical screening platform for comprehensive toxicity prediction.
Why ToxCML?
Rapid Screening
Efficiently evaluates thousands of chemical compounds.
Cost Reduction
Minimizes reliance on expensive in vivo or in vitro testing.
Ethical Compliance
Reduces the need for animal testing.
Resource Efficiency
Prioritizes high-potential candidates for experimental follow-up.
Comprehensive Risk Assessment
Predicts a wide array of toxicological endpoints with explainable AI.
Technical Highlights
General Workflow of ToxCML
1. Data Input
- Accepts SMILES notation for chemical structures.
- Visualizes chemical structures directly within the platform.
2. Feature Extraction
- For QSAR (Machine Learning) Models:
- Fingerprints: Morgan Circular, MACCS Keys, Atom Pair Fingerprint (APF)
- Descriptors: Physicochemical properties
- For Read-Across Models:
- Fingerprints: Morgan Circular, MACCS Keys, Atom Pair Fingerprint (APF), RDKit Fingerprint
3. Machine Learning & Read-Across
- Machine Learning Models: Random Forest, XGBoost, and SVM (QSAR-based AI)
- Read-Across Approach: Tanimoto similarity to predict toxicity from structurally similar compounds
- Hybrid mfCoQ-RASAR Prediction: Integrates ML (QSAR) and read-across outputs for robust, explainable, and reliable predictions
4. Evaluation & Prediction
Toxicity Endpoints (18):
- Eye Irritation
- Skin Irritation
- Carcinogenicity
- Acute Oral Toxicity
- Acute Dermal Toxicity
- Acute Inhalation Toxicity
- FDA Maximum Daily Dose (FDA MDD)
- Hematotoxicity
- Mitochondrial Toxicity
- Skin Sensitization
- Respiratory Toxicity
- AMES Mutagenicity
- Drug-Induced Liver Toxicity (DILI)
- Hepatotoxicity
- Developmental Toxicity
- Cardiotoxicity
- Drug-Induced Nephrotoxicity (DIN)
- Neurotoxicity
- Applicability Domain Analysis: Evaluates whether each compound is sufficiently represented within the model’s chemical space to support reliable prediction.
- Fragment Contribution Map: Provides interpretable visual insight into the molecular fragments driving predicted toxic or non-toxic outcomes.
- Exportable Results: Supports seamless download of prediction outputs for downstream analysis, reporting, and documentation.
