Predictive toxicology
AI/ML models for acute, chronic and endpoint-specific toxicity, supported by molecular fingerprints, physicochemical descriptors and curated chemical data.
- QSAR/QSPR modeling
- Read-across support
- Applicability-domain analysis
AI-driven · in silico · in vitro · human-relevant
From molecular predictions to biological evidence.
PharmaToxAI integrates computational toxicology, ADMET modeling and human cellular models to support earlier, better-informed decisions on compounds and ingredients.
The problem
Prediction scales. Experiments anchor reality. The useful system connects both.
Computational models can screen large chemical spaces quickly, but confidence drops outside their domain. Experimental systems provide biological evidence, but cannot be applied indiscriminately to every candidate. PharmaToxAI is built around the feedback loop between them.
Solutions
Early-stage evidence for compound decisions.
Modular analyses can be used independently or combined into an integrated evaluation workflow.
ADMET & barrier prediction
Early characterization of pharmacokinetic behavior, bioavailability and permeability to biologically relevant barriers.
- PBPK / ADME simulations
- Blood-brain barrier
- Dermal and placental permeability
Human cellular validation
Experimental evaluation in human cell systems when computational predictions require biological confirmation.
- 2D cellular models
- 3D cellular models
- Viability and functional endpoints
Compound prioritization
Integrated computational assessment to rank candidates before committing resources to more expensive experimental stages.
- Lead prioritization
- Molecular modeling
- Go / No-Go support
Workflow
A closed loop from structure to evidence.
01
Define
Compound, ingredient, formulation or candidate is characterized together with the decision that needs to be made.
→
02
Predict
AI/ML, QSAR/QSPR, molecular modeling and PBPK/ADME methods estimate safety and biological behavior.
→
03
Validate
Selected endpoints are tested in human 2D/3D cellular systems where experimental confirmation adds value.
→
04
Integrate
Computational and experimental evidence are interpreted together in a consolidated technical report.
Technology
Methods selected for the question, not for the buzzword.
The core pipeline uses established machine-learning approaches and rigorous validation, with deeper models or mechanistic simulations added when the endpoint justifies them.
Machine learningXGBoost · LightGBM · Random Forest
Molecular representationFingerprints · physicochemical descriptors
Model validationScaffold split · external validation
CheminformaticsQSAR · QSPR · chemical-space analysis
PharmacokineticsPBPK · ADME simulations
Experimental modelsHuman cell systems · 2D · 3D
Scientific validation
Models are useful only when we know where they fail.
PharmaToxAI emphasizes validation strategy as strongly as algorithm selection: scaffold-aware splits, independent external validation and iterative comparison against experimental data are used to reduce chemical memorization and expose model limits.
Applicability matters. Predictions are interpreted in the context of the chemistry represented during training.
Experimental feedback matters. In selected endpoints, predictions can be tested in cells and used to refine subsequent modeling.
Decision context matters. Outputs are structured to support early prioritization rather than replace downstream evidence.
Applications
Built for pharmaceutical, cosmetic and biomedical development.
Pharma
Candidate prioritization, safety profiling, ADMET characterization and preclinical decision support.
Cosmetics
Ingredient safety, dermal compatibility and animal-reduction strategies supported by predictive and human-cell methods.
Biomedical R&D
Custom computational and experimental studies for compounds, formulations and mechanistic questions.
Team
Interdisciplinary by construction.
PharmaToxAI combines computational modeling with experimental physiology, cell biology and human-relevant models at IFIBIO Houssay (UBA–CONICET).
JC
Dr. Juan J. Casal
Computational modeling · bioinformatics · AI
Pharmacist and PhD in Pharmacy. Computational drug design, molecular dynamics, scientific data analysis and machine learning.
GD
Dr. Gisela Di Giusto
Cell physiology · aquaporins · experimental models
CONICET Adjunct Researcher. Cellular and molecular pathophysiology, renal physiology, cell migration and 2D/3D experimental systems.
VR
Dr. Valeria Rivarola
Renal cell physiology · cancer biology · 3D models
CONICET Adjunct Researcher. Cellular physiology, acid-base regulation, aquaporins, renal cancer and experimental cell systems.
NM
Dr. Nora Alicia Martínez
Placental biology · vascular development · human cell models
CONICET Adjunct Researcher. Molecular mechanisms of human placental vasculature development and its dysregulation in preeclampsia.
Contact
Have a compound or development question?
Tell us what decision you need to make and what data you already have. We can define the appropriate computational, experimental or integrated evaluation strategy.
Contact PharmaToxAI
gdigiusto@fmed.uba.ar
PharmaToxAI on LinkedIn
IFIBIO Houssay · UBA–CONICET · Buenos Aires, Argentina