What Models to Engage

Data De-risking Sprint (30–60 days): causal/graph analytics + patient stratification → ranked targets & biomarkers.
Target to Lead Studio (12–24 weeks): foundation model design + ADMET + in vivo.
Adaptive Trial Lab: Digital twins, Bayesian platform designs, EHR enabled endpoints.

Why UAB and SPARC

10+

AI Drug Discovery in pipeline

160+

Invited Talks

150+

Publications

40+

Academic and industrial partners

180+

Platform Users

Multi-scale Omics integration for AI model training: gene, protein, pathway, cell, tissue, organ, body, cohort, and population
Text to knowledge graphs for literature & EHR notes: e.g., cTAKES/medspaCy + LlamaIndex/RAG over curated corpora
In-house developed proven tools of knowledge graphs & causal inference:
- BEERE – biomedical entities expanded, ranked, and visualized
- BioRSP – spatial pattern and biomarker identification
- GeneTerrain – gene-disease-drug network
- HAPPI-2 – 3M+ protein–protein interactions for target mapping
- PAGER – curated pathway and gene signature resource for precision biology
- PETS – multi-scale therapeutic effect and toxicity simulation
- WINNER – biomolecular characterization and prioritization
- WIPER – biomolecular association networks

Agentic AI with reinforcement learning to leverage ESMFold, Boltz-2, equivariant GNN, SchNet/DimeNet++/PaiNN, RDKit+DeepChem, and in-house fine-tuned models to navigate the vast chemical space to find optimal drug candidates (selectivity, ADME, and safety)
Generative AI (GANs, VAEs, REINVENT4, AiZynthFinder, etc.) coupled with multi-objective optimization & active learning (Ax/BoTorch, Nevergrad, etc.) to create novel synthesizable molecules from scratch
Causal AI (roadrunner, causallib, etc.) for cause-and-effect relationships beyond simple statistic correlation, a step for drug candidate nomination based on the true mechanism of action to ensure clinical success
Wet-lab validation: synthesis, in vitro, and in vivo
Case studies
- Skin fibrosis inhibitors (16 weeks) – gene-analysis → virtual screenings of multi drug modalities → in vitro validation → in vivo validation ongoing
- Alzheimer’s Disease (3 months) – paper on new target PHGDH from literature → AI model development → hit discovery and analog design → patent filing
- Viral inhibitors (12 weeks) – market analysis to determine antiviral spectrum → virtual screenings of viral and host protein dual inhibitors → in vitro validation → hit-to-lead optimization ongoing

Adaptive trial design & digital twins: Bayesian platform trials (arm dropping/addition), causal forests for subgroup detection, emulators for in‑silico trials
QSP/PK‑PD: Bayesian hierarchical models (e.g., PK‑Sim) and integrated ML predictions
EHR/RWD pipelines: OMOP CDM + OHDSI ATLAS, FHIR APIs, differential privacy (Opacus), and federated learning (Flower/FedML) for multi‑site studies
UAB unique resources:
- Patient scale & diversity – 35% black patients (vs ~25% U.S. hospital average), high-acuity cases from Alabama’s only Level I trauma center
- Specialized disease cohorts – nationally recognized centers (Alzheimer’s, Sickle Cell, Cancer, Neurodegeneration, Pulmonary Fibrosis, Rare Diseases), longitudinal multi-modal data, ideal populations (for studying disease progression, comorbidities, drug effects)
- Clinical research powerhouse – NIH-funded centers with harmonized national datasets (e.g., i2b2 with 9+ billion facts on 1+ million patients)
- Advanced AI integration – causal AI, multimodal deep learning, digital twin simulations, secure on-prem computing
- Trusted data governance – reproducibility (automated lineage, audit trails, etc.) and compliance (PHI de‑identification, access control, 21 CFR Part 11 REDCap systems)